JP6059957B2 - Wildlife identification device, wildlife identification method, and program - Google Patents

Description

  The present invention relates to a wildlife identification device, a wildlife identification method, and a program, and more particularly, to a wildlife identification device, a wildlife identification method, and a program that can identify a wildlife that is flapping.

  In recent years, the number of birds such as pigeons and crows has increased in urban areas and residential areas, and bird damage such as excrement damage, noise damage, and pathogen transmission has become a problem. In order to prevent such bird damage, it is first necessary to detect the presence of a bird.

  Patent Document 1 states that “a microphone that captures the cry of a bird that is stopped on or near the electric wire, and a cry recognition device that recognizes the cry of a bird that should be removed from the sound captured by the microphone. (See claim 1 of Patent Document 1).

  Patent Document 2 states that “image data obtained by photographing a stationary harmful bird for a predetermined period of time is subjected to an erasing process for erasing moving objects to obtain target data. The reference data specifying the mode and the target data are collated, and the number of the harmful birds in the stopped state and the stop position are determined by confirming whether the target data has the same or similar mode as the reference data. “A method for determining the number of harmful birds and the stopping position” (refer to claim 1 of Patent Document 2).

  Patent Document 3 discloses a “personal authentication device” that identifies and recognizes a person's face from moving image data.

JP 2009-207438 A JP 2011-138310 A JP 2012-123676 A

  Patent Document 1 is an invention about a call recognition device that recognizes using a bird's cry as a recognition target. When a bird / beast as a recognition target does not sing or when it is difficult to acquire sound due to bad weather, etc. Is unrecognizable.

  Further, Patent Document 2 is a method capable of recognizing only a bird and animal parked on an electric wire or the like using so-called template matching, and cannot recognize a bird and animal in flight. For this reason, it cannot be recognized until the birds and beasts stop, and cannot be used when the real time is required for the identification of the birds and beasts or when it is necessary to recognize the birds and beasts in flight or walking.

  The apparatus for performing face identification exemplified in Patent Document 3 and the like is not suitable for recognizing birds and beasts because it is for human face authentication. In addition, even if the face of a beast is recognized using data for birds and beasts, it is not always the case that the face is facing the front on the screen, and in that case, the recognition is impossible. .

  In order to prevent bird damage, it is effective to identify the presence of a bird even when a human is absent, and to appropriately take measures such as bird removal and prevention of bird entry. Furthermore, rather than identifying birds that are parked on wires, roofs, and ground, if you can identify a flapping bird just before landing, you can quickly take action such as bird removal before or after landing. It is more effective because it can be executed immediately after.

  The problem to be solved by the present invention is to provide a wildlife identification apparatus, a wildlife identification method, and a program capable of identifying a wildlife that is flapping among wildlife present in a specific spatial region.

Means for solving the problems are as follows:
(1) For n images P _t (t is an integer from 1 to n) acquired by shooting a specific spatial region, the image P _t and the image P _t shot before the image P _{t are used. -r (r} represents. an integer of 1 to t-1) performs the difference processing between each of the at least one image P _t-r of the inter-frame difference processing unit that acquires a difference image between frames,
A first divided region acquisition unit that divides the inter-frame difference image into m regions and acquires m regions R _s (s represents an integer of 1 to m);
The region R _s is divided into Assuming wing area as the wings of the assumed body region and wildlife as fuselage wildlife, the magnitude of the respective body region and the wings region, the number of pixels obtained by the differential processing A flying bird identification unit that determines whether or not the region R _s is a first wildlife region where one fluttering wildlife exists;
It is a birds and beasts identification apparatus provided with.
A preferred embodiment of (1) is as follows:
(2) the flying wildlife identification unit, three were divided into regions, the central region of the body region of which adjacent the region R _s in the horizontal direction, a region on both sides as the wing area, the body region smaller than the number of pixels preset value, and, at least when one number of pixels is greater than a preset value, the first wildlife there wildlife that fluttered with one bird the region R _s is present in the blade region It is determined as an area.
(3) In the wildlife identification device of (1) or (2), performs differential processing of the background image that has been captured in advance for the image P _t and the specific spatial regions, acquires a background difference image A background difference processing unit;
A second divided region acquisition unit that divides the background difference image into x regions and acquires x regions R _y (y is an integer of 1 to x);
A stationary walking animal identification unit that determines whether or not the region _Ry is a second wildlife region where one stationary or walking wildlife exists;
Is provided.
(4) In the wildlife identification device of (3), using the determination result of the image P _t-r in the flying birds and beasts identification unit and / or the retention walking wildlife identification section, of the region R _s and the region R _y Each includes a wildlife identification correcting unit that determines whether the first wildlife area, the second wildlife area, or the non-existent wildlife area.
(5) In the bird and beast identification apparatus according to any one of (1) to (4), a flying bird and animal counting unit that counts the number of the first birds and animals existing region as the number of birds and animals in a state of flapping,
A stationary walking wildlife counting unit that counts the number of the second wildlife existing area as the number of stationary or walking wildlife;
A total bird count unit for obtaining a total number of birds and beasts that is the sum of the number of birds and birds that are flapping and the number of birds and animals that are stopped or walking;
If there is an overlapping area between the first wildlife area and the second wildlife area, it is considered that the area is double-counted, and the total number of birds and the number of birds in the flapping state or the stop A bird count correction unit that subtracts the count from the number of birds in the middle or walking, and obtains the number of birds in the flapping state, the number of birds in suspension or walking, and the total number of birds and animals. .

Another means for solving the problem is as follows:
(6) For n images P _t (t is an integer from 1 to n) acquired by shooting a specific spatial region, the image P _t and the image P _t shot before the image P _{t are used. -r (r} represents. an integer of 1 to t-1) performs the difference processing between each of the at least one image P _t-r of, obtaining a difference image between frames,
Dividing the inter-frame difference image into m regions to obtain m regions R _s (s is an integer from 1 to m);
The region R _s is divided into Assuming wing area as the wings of the assumed body region and wildlife as fuselage wildlife, the magnitude of the respective body region and the wings region, the number of pixels obtained by the differential processing And determining whether or not the region R _s is a first region where a single flapping bird is present, and
A method for identifying birds and beasts.

Further, another means for solving the problem is as follows:
For acquired by photographing a specific spatial region n images P _{t (t} is an integer of 1 to n.), Images were taken before the image P _t and the image P _t P _t-r ( r represents an integer of 1 to t−1), and performs a difference process with each of at least one image P _t−r to obtain an inter-frame difference image;
Dividing the inter-frame difference image into m regions to obtain m regions R _s (s is an integer from 1 to m);
The region R _s is divided into Assuming wing area as the wings of the assumed body region and wildlife as fuselage wildlife, the magnitude of the respective body region and the wings region, the number of pixels obtained by the differential processing And determining whether or not the region R _s is a first region where a single flapping bird is present, and
A program that causes a computer to execute.

  According to the present invention, it is possible to provide a wildlife identification device, a wildlife identification method, and a program that can identify a wildlife that is flapping. In order to prevent damage from birds and beasts, it is desirable to be able to quickly detect birds and beasts that are approaching. According to the present invention, when determining the presence or absence of flapping birds and beasts for a certain image, an image taken in the past from that image is used, and an image taken after that image is not used. Processing for making the determination can be performed while photographing with a camera or the like, and the determination result can be acquired in real time. Rather than identifying birds and beasts that are parked on the wire, roof, ground, etc., if the birds and birds that fluttered just before landing can be identified, measures such as threatening and driving away the birds and beasts can be performed. Since it can be executed immediately before landing or after landing, for example, birds and beasts that damage humans can be surely removed. Also, according to the present invention, since the flapping birds and beasts are identified, for example, moving objects such as airplanes and gliding birds invade into the space area where the birds and beasts to be identified are photographed. In this case, it is possible to suppress the possibility of misrecognizing it as an identification target bird and beast, and to reliably identify the identification target bird and beast.

  According to the present invention, it is possible to provide a wildlife identification device that can separately identify a wildlife that is flapping and a wildlife that is stopped or walking.

  According to the present invention, since the wildlife identification correction unit is further provided, it is possible to provide a wildlife identification apparatus that can accurately identify a wildlife that is flapping and a wildlife that is stopped or walking.

  According to the present invention, since it further comprises a bird counting unit, it provides a bird identification device capable of acquiring the number of birds that are flapping, the number of birds that are parked or walking, and the total number of birds and animals that are the total of these. can do.

FIG. 1 is a block diagram showing a schematic configuration of a wildlife identification apparatus which is an embodiment of the wildlife identification apparatus of the present invention. FIG. 2 is a diagram illustrating a connection state between the bird and animal identification apparatus illustrated in FIG. 1 and an imaging unit. FIG. 3 is a block diagram showing a processing unit realized by executing a program by the CPU in FIG. FIG. 4 is a flow chart showing an embodiment of the method for identifying birds and beasts of the present invention. FIG. 5 is a diagram showing an embodiment of the inter-frame difference image acquired by the difference process. 5 (a) is an image _{P t} is the differential processing target, FIG. 5 (b) is an image _{P t-1} of one frame before the image _{P t} is a target of the difference processing, FIG. 5 ( c) is an inter-frame difference image after the difference processing is performed on the image P _t and the image P _t−1 . It is a conceptual diagram for demonstrating the state which divided | segmented the inter-frame difference image shown to Fig.6 (a) and FIG.5 (c). FIG. 6B is a conceptual diagram for explaining an example of a method for selecting the first inspection region from the inter-frame difference image shown in FIG. FIG. 7 is an explanatory diagram for explaining a state when the first inspection region is divided into three regions. FIG. 8 is a diagram illustrating an embodiment of a background difference image acquired by difference processing. Figure 8 (a) is a background image, Fig. 8 (b), an image P _t is the differential processing target, FIG. 8 (c), a background differential image after difference processing. FIG. 9 is an explanatory diagram for showing a difference identification area to be subjected to difference processing. FIG. 10 is a reference color histogram created for a stationary pigeon. FIG. 11 is an explanatory diagram showing an outline of a process for performing template matching by converting a template image and a second inspection region into a binary image. FIG. 12 is an explanatory diagram for explaining the transition state probability.

  An embodiment of a bird and animal identification apparatus according to the present invention will be described below with reference to the accompanying drawings.

  FIG. 1 is a block diagram showing a schematic configuration of a wildlife identification apparatus which is an embodiment of the wildlife identification apparatus of the present invention. The bird and animal identification apparatus 1 has a built-in computer, and executes various programs to perform calculations and processing. For example, a personal computer is used as the computer.

  The computer includes a CPU (Central Processing Unit) 3, a ROM (Read Only Memory) 4, and a RAM (Random Access Memory) 5, which are connected to each other via a bus or a controller chip. Further, the CPU 3, the ROM 4, and the RAM 5 are connected to the storage device 6, the input device 7, and the display device 8 through a bus, a control chip, an interface, and the like.

  The ROM 4 is a nonvolatile memory that stores programs and data in advance. The RAM 5 is a memory that temporarily stores the program and data when the program is executed. The CPU 3 is an arithmetic processing unit that executes processing described in the program. The CPU 3 loads a program stored in the ROM 4 or the storage device 6 to the RAM 5 and executes it. Thus, a processing unit that executes processing described in the program is realized.

  The storage device 6 is a device having a recording medium that stores an operating system (not shown), various programs 61, and various data 62 such as an image captured by the imaging unit 2 described later. The storage device 6 is preferably capable of storing a plurality of images, and a non-volatile memory, a hard disk drive device, a disk array device, or the like is used. The program 61 may be recorded on a portable recording medium such as a CD-ROM or DVD-ROM, and installed in the bird and animal identification apparatus 1 from the recording medium.

  The display device 8 is a device that displays an operation screen or the like. For example, a liquid crystal display is used as the display device 8. The input device 7 is a device for inputting an operation on the operation screen. For example, a keyboard or a mouse is used as the input device 7.

  As shown in FIG. 2, the bird and animal identification apparatus 1 of this embodiment is connected to the imaging unit 2 by wire or wirelessly. When the imaging unit 2 is connected to the bird and animal identification device 1, an image obtained from the imaging unit 2 such as a camera can be stored in the RAM 5 or the storage unit 6 and processing described later can be performed in real time. The image pickup unit 2 may not be connected to the bird and animal identification device 1, and in that case, an image taken by the image pickup unit 2 is recorded on a portable recording medium such as a CD-ROM or a DVD-ROM. Then, the recording medium may be connected to the wildlife identification device 1 to read out an image from the recording medium for performing the wildlife identification processing described later.

  The imaging unit 2 only needs to be able to photograph the birds and beasts that have entered the installation area of the imaging unit 2, and specifically, a camera capable of capturing still images and / or moving images can be employed. As the imaging mode of the imaging unit 2, a mode in which the imaging state is constantly maintained, a mode in which imaging is regularly performed, and a moving sensor with a motion sensor (not shown) attached to the imaging unit 2 sense a moving object. For example, the imaging unit 2 may start imaging when the imaging unit 2 starts. The image to be stored may be either a color image or a black and white image.

  Examples of the birds and beasts whose presence or absence can be identified by the bird and animal identification device 1 include creatures having wings and flapping during flight, such as pigeons, crows, sparrows, starlings, bats, etc. Can be mentioned. Among these, pigeons and crows can be given as preferable examples as birds and beasts that are not under human control and may be harmful birds such as feces or pathogens. In addition, the creatures that can be identified by the bird and animal identification device 1 are not particularly limited to birds and beasts, and by appropriately setting the imaging region of the imaging unit 2 according to the size of the organism to be identified, insects having wings Etc. can also be identified.

  As the installation location of the imaging unit 2 in the bird and beast identification apparatus 1, for example, an area that needs to be managed by a human and that has a time zone in which the human is absent can be cited. Examples include verandas, outdoor food storage areas, food storage areas, garbage dumping areas, fields, and building storage areas. In addition, since the bird and beast identification apparatus of the present invention can identify the flapping birds and beasts just before landing, it is preferably installed so that the place where the birds and beasts may land can be photographed. .

  FIG. 3 is a block diagram showing a processing unit realized by executing a program by the CPU in FIG. The processing unit 9 is (1) a flying wildlife processing unit 11 for identifying a flying wildlife, an inter-frame difference processing unit 12, a first divided region acquisition unit 13, and a flying wildlife identification processing unit 14, 2) A background difference processing unit 22, a second divided region acquisition unit 23, and a stationary walking wildlife identification unit 24, which are a stationary walking wildlife processing unit 21 for identifying a stationary or walking wildlife, and (3) flying The determination result processing unit 31 is a determination result processing unit 31 that performs processing on the determination result of identifying the inside wild bird and / or the determination result of identifying the stationary or walking bird and beast.

  The bird and animal identification apparatus of the present invention only needs to have at least the flying bird and animal processing unit 11 shown in the above (1), and by having the flying bird and animal processing unit 11 shown in the above (1), it flutters immediately before landing. Can identify state birds and beasts. The stationary walking birds and beasts processing unit 21 shown in the above (2) is not indispensable, but by having this processing unit, it is possible to identify not only the birds and birds that are flapping but also all the birds and animals that are present on the screen. . In addition, the determination result processing unit 31 shown in (3) is not essential, but by having this processing unit, it is possible to improve the identification accuracy of birds and birds that are flapping and / or birds and animals that are stopped or walking. The determination result according to (1) and / or (2) can be processed as required, such as displaying the number of birds and beasts, saving the determination result, and transmitting the determination result to another device.

  The bird and animal identification apparatus of the present invention includes an aspect having only the processing section shown in (1), an aspect having the processing section shown in (1) and (2), and the processing shown in (1) and (3). The aspect which has a process part which has a process part shown to said (1)-said (3) can be mentioned.

  Next, each of the processing units shown in (1) to (3) will be described in detail below.

The inter-frame difference processing unit 12 performs an image P _t and an image P _{t− on} n images P _t (t is an integer from 1 to n) acquired by photographing a specific space area with a camera or the like. Difference processing with ₁ is performed, and an inter-frame difference image is acquired. Here, assuming that there are n images P _t (t is an integer of 1 to n) in one continuous shooting, the images P ₁ , P ₂ ,. Called P _t ... P _n . Note that the time interval between the image P _t and the image P _t−1 is appropriately set according to the number of flappings per predetermined time of the target animal and animal to be identified, the data processing speed, and the like. If an image is acquired at an interval of 15 to 60 sheets / second and a process described later is performed, the presence or absence of a pigeon can be accurately identified. Image P _t may be cut at a predetermined time interval as a frame from a moving image, it may be selected by continuously or at predetermined intervals among the continuous shooting still images.

From the inter-frame difference image obtained by the difference process, a region having a change between the image P _t and the image P _t−1 is extracted. For example, when the image P _t and the image P _t−1 are represented by RGB values for each pixel, and the difference value of the RGB value for each corresponding pixel is equal to or greater than a preset threshold value, the coordinates of the pixel change. It is saved as a certain area. 5 (a) is an image _{P t} is the differential processing target, FIG. 5 (b) is an image _{P t-1} of one frame before the image _{P t} is a target of the difference processing, FIG. 5 ( c) is an inter-frame difference image after the difference processing is performed on the image P _t and the image P _t−1 . For example, as shown in FIG. 5 (a) and (b), the pigeons that is reflected in the image P _t, there is a movement from the image P _t-1 to the position and orientation or the like of the pigeon between the image P _t In that case, the pigeon is extracted and shown as a white part in FIG. The background without motion is not extracted and is shown as a black part in FIG.

  Further, the inter-frame difference processing unit 12 may perform the following post-processing after performing the difference processing. The post-processing is noise removal processing for removing noise such as isolated points from a frame difference image using a noise removal filter such as a median filter. Since isolated points that are smaller than the area expected for the size of the birds and beasts are unlikely to be birds and beasts, removing such isolated points in the inter-frame difference image will increase the data processing speed thereafter. In addition, the presence or absence of birds and beasts can be accurately identified.

The first divided region acquisition unit 13 divides the inter-frame difference image into m regions, and acquires m regions R _s (s indicates an integer from 1 to m). FIG. 6A is a conceptual diagram for explaining a state in which the inter-frame difference image shown in FIG. 5C is divided. In the inter-frame difference image shown in FIG. 6, the inter-frame difference image is divided into 5 vertical columns and 5 horizontal rows, and is divided into 25 rectangular regions in total. At this time, the size and shape of the region R _s are not particularly limited, but if the size and shape including the flapping birds and beasts are set in advance, the presence or absence of the pigeon can be accurately identified. Can do.

In the first divided region acquisition unit 13, for each of the regions R _s , when the number of pixels in which the pixel value obtained by the difference processing is equal to or greater than a preset threshold value has a value equal to or greater than a preset threshold value, described later. It is preferable to perform a first inspection region selection process that selects the first inspection region R _sy to be processed by the flying bird / animal identification unit 14. When the number of pixels in the region R _s is equal to or greater than a preset threshold value, there is a possibility that there is a fluttering bird and beast, so the first inspection region R _sy is selected. When the number of pixels in the region R _s is less than a preset threshold value, there is a low possibility that there is a moving bird and beast in the region R _s , so this is set as a non-inspection region R _sn , and the process ends here. The first examination area selection process is not essential, but by performing this process, it is possible to remove areas where there is a low possibility of the existence of birds and beasts, so that the data processing speed thereafter can be increased.

The method for selecting the first inspection area is not limited to the above-described method except for an area where there is a high possibility that there is no bird in the state of flapping due to the size of the number of pixels. For example, in addition to or instead of the above-described method Te, by treatment with a color histogram or template matching exemplified stationary walking wildlife processor to be described later, determines the presence of the wings of birds and animals that are fluttered for each region R _s, the wildlife of fluttered by that condition exists Only when it is determined, the process may proceed to the next process.

The method of acquiring the first inspection region R _sy in the first divided region acquisition unit 13 is not limited to the method described above, and an area having a predetermined size is displayed on the inter-frame difference image as shown in FIG. , And try all combinations that can be placed on the inter-frame difference image, and examine the number of pixels whose pixel values obtained by the difference processing in each region are greater than or equal to a preset threshold value. It may be determined that there is a possibility that there is a bird or beast flapping the region R _s where the number of pixels is equal to or greater than a preset threshold value, and may be selected as the first inspection region R _sy .

The flying birds and beasts identification unit 14 uses the region R _s or the first inspection region R _sy (hereinafter, the first inspection region R _sy will be described) as the body of the bird and beast and the wings of the bird and beast. The first inspection region R _sy is divided into hypothetical wing regions, and the magnitude relationship between the fuselage region and each of the wing regions is compared with a predetermined threshold value of the number of pixels obtained by the difference processing. It is determined whether or not this is the first wildlife region where there is one flapping bird. When the number of pixels is small, it indicates that the area is poorly changed, and the area may be a torso body. When the number of pixels is large, it indicates that the region changes rapidly, and the region may be a wing part of a bird or beast. When the body region is a poorly changing region and the wing region is a rapidly changing region, the first inspection region _Rsy is a first wildlife region where a single flapping bird is present. Judge that there is.

As a method for dividing the first inspection region R _sy , various methods can be adopted as long as the first inspection region R _sy can be divided into a torso and a wing part. For example, the first inspection region R _sy is _arranged in the horizontal direction. Can be divided into three regions adjacent to each other. FIG. 7 is an explanatory diagram for explaining a state when the first inspection region is divided into three regions. Note that the first inspection area shown in FIG. 7 is an image obtained by dividing an image of a photographed pigeon into m pieces without performing the difference processing for easy understanding. It is not a processed image. In this method, it is assumed that the central region R _syb of the three regions is a fuselage region, and regions R _sya and R _syc on both sides _thereof are wing regions. When the number of pixels in the central region R _syb is smaller than a preset threshold value and the number of pixels in at least one of the regions R _sya and R _syc on both sides thereof is larger than the preset threshold value, the central region R _syb is defined as the body region. _Assuming that the regions R _sya and R _syc on both sides are wing regions are correct, it is determined that the first inspection region R _sy is a first wildlife region where one fluttering wildlife exists. When the number of pixels in the central region R _sb is greater than or equal to a preset threshold value, or when the number of pixels in the regions R _sa and R _sc on both sides thereof is less than or equal to a preset threshold value, the assumption is denied and this It is determined that there is no flapping bird in the first inspection region R _sy . Note that the threshold values for the number of pixels in the central region R _syb and the regions R _sya and R _syc on both sides thereof may be the same or different.

As another method of dividing the first inspection region R _sy , an image of a bird and beast to be identified is acquired in advance, and the substantially circular trunk of the bird and beast is used as a template image, and the first inspection region R is obtained using this template image. _A template matching can be performed for all areas of _sy , and when there is a substantially circular area that matches the template image, a method of dividing the area into two areas, that is, the approximately circular area and the other areas can be cited. . In this method, a substantially circular region is assumed to be a fuselage region, and other regions are assumed to be wing regions. When the number of pixels in a substantially circular area is smaller than a preset threshold value and the number of pixels in other areas is larger than a preset threshold value, the substantially circular area is defined as a trunk area and the other areas as wing areas. Assuming that the assumption is correct, it is determined that the first inspection region R _sy is a first wildlife region where one fluttering wildlife exists. When the number of pixels in the substantially circular area is equal to or greater than a preset threshold value, or when the number of pixels in other areas is equal to or less than the preset threshold value, the assumption is denied and the first inspection area R _sy It is determined that there is no flapping bird. In the case where substantially circular area that matches the template image does not exist, it is impossible to divide the first inspection region R _sy the fuselage region and wings region, in the first inspection region R _sy 1 Since there is a high possibility that there are no flapping birds and beasts, it is determined that there is no flapping bird in the first inspection region _Rsy .

The following method can be given as another method for dividing the first inspection region R _sy . First, images P _{t -1} , P _t-2 ... P _t-r (r is an integer from 1 to _t−1) of a plurality of frames taken before the image P _t and the image P _t . Frame difference images Q _t−1 , Q _t−2 ,... Q _t−r are created by difference processing with each. In each of the inter-frame difference image Q _t-r, given the size of the area scanned on the inter-frame difference image Q _t-r will, set pixel value obtained by the subtraction processing for each scanning regions in advance The number of pixels that are equal to or greater than the threshold value is checked, and an area in which the number of pixels is in a predetermined range is selected as an estimated body area. If there is at least one area selected as the estimated body area for a specific area in the frame difference images Q _{t−1 to} Q _t−r , the number of pixels for that area is set to the image Q _t−1 to The total number of pixels is calculated by adding _r of Q _t−r . The first inspection region R _sy is divided into these two regions, assuming that the region in which the total number of pixels is within a predetermined range is a fuselage region and the other region is a wing region. First, when the body region does not exist, it is determined that there is no flapping bird in the first inspection region _Rsy . When the trunk region is present in the first inspection region R _sy , the number of pixels obtained by the difference process between the image P _t and the image P _t−1 in the trunk region is smaller than a preset threshold value, and When the number of pixels obtained by the difference process between the image P _t and the image P _t−1 in the other region is larger than a preset threshold value, the region obtained based on the estimated body region is the body region, and the others The first inspection area R _sy is determined to be a first wildlife area where one fluttering wildlife exists. When the number of pixels in the body region is greater than or equal to a preset threshold value, or when the number of pixels in other regions is less than or equal to a preset threshold value, the above assumption is denied, and there is one bird in this region R _sy. It is determined that there are no flapping birds and beasts.

In the first inspection region R _sy , the fuselage region and the wing region are assumed, and as a method for determining whether this assumption is correct, as described above, the determination is based on the size of the number of pixels obtained by the difference processing. However, the determination may be made as follows. First, a plurality of images of the flapping state of the birds and beasts to be identified are acquired in advance, and the average value of the RGB density of the wings is calculated. In the first inspection region R _sy, for example by any of the dividing method described above, assuming a certain region the blade area, the assumed area as wing area, the image P _1, P ₂ ··· P _t each color space , The region where the distance from the average value of RGB density acquired in advance is equal to or less than a predetermined value is extracted. Using this as an extraction region, binarization processing is performed so as to distinguish the extraction region from other regions, and a binarization processing image in a region assumed as a wing region is created. Next, the differential processing with the immediately preceding image in the order as for the binarized image image _{P 2} and the image _{P 1,} the image _{P 3} and the image _P 2 · · · image _{P t} and the image _{P t-1} , Frame difference images T ₂ , T ₃ ... T _t are acquired. The total number of pixels present in each frame difference image T _t is calculated, and if the total number of pixels is equal to or greater than a preset threshold value, the assumption that a certain region is a wing region is correct and the first The inspection area R _sy is determined to be the first wild beast existence area in which one fluttering wild beast exists. If the total number of pixels in the region assumed as the wing region is less than a preset threshold, the assumption is denied and it is determined that there is no flapping bird in this region R _sy. .

As described above, after determining that the first inspection region R _sy is the first wildlife region where one fluttering wildlife exists in the flying wildlife identification unit 14, a process using a color histogram described later, and a template By performing verification processing such as processing using matching, or processing combining these, verification of the determination result in which the first inspection region R _sy is determined to be the first wildlife region may be performed. The result of this verification process, the determination result if the result of wildlife that fluttered with one bird in the first inspection region R _sy is absent is obtained, in which a first inspection region R _sy is judged that the first wildlife existing region And it is determined that there is no flapping bird in the first inspection region R _sy .

The background difference processing unit 22 previously captures n images P _t (t represents an integer of 1 to n) acquired by capturing a specific spatial region with a camera or the like and the specific spatial region. The difference process with the reference background image acquired in this way is performed, and the background difference image is acquired. For how to obtain a picture P _t, it is the same as the inter-frame difference processing unit 12 described above. In the background difference image, different areas are extracted between the image _Pt and the background image. For example, when the image _Pt and the background image are represented by RGB values for each pixel and the difference value of the RGB values for each corresponding pixel is equal to or greater than a preset threshold, the coordinates of the pixels are the image P and the background image. Are saved as different areas. Figure 8 (a) is a background image, Fig. 8 (b), an image P _t is the differential processing target, FIG. 8 (c), a background differential image after difference processing. As shown in FIG. 8C, after the difference processing, a pigeon that is stopped or walking on the roof is extracted and shown as a white portion. A region that is the same as the background image is not extracted and is shown as a black portion in FIG.

  The background difference processing unit 22 may perform the following preprocessing before performing the difference processing. In the background image, the pre-processing is a process in which a region in which a bird or beast is likely to stop or walk is set in advance, and this region is set as a difference identification region as a difference processing target. Examples of the area where birds and beasts may stop or walk include the roof, the ground, and the periphery of electric wires. For example, as shown in FIG. 9, a region including a bird or animal that is parked or walking on the roof in the background image shown in FIG. In FIG. 9, the difference identification area is shown in white. By setting the differential identification area, it is possible to prevent the birds and animals in flight from being erroneously extracted, and it is possible to accurately identify the presence or absence of birds and animals that are stopping or walking. Moreover, since the area | region where the birds and beasts which are stopping or walking does not exist can be removed beforehand, the data processing speed after this can be raised.

  The background difference processing unit 22 may perform the following post-processing after performing the difference processing. The post-processing is noise removal processing for removing noise such as isolated points from the background difference image using a noise removal filter such as a median filter. An isolated point smaller than the area assumed as the size of the bird and beast is unlikely to be a bird and beast. Therefore, by removing such an isolated point from the background differential image, the subsequent data processing speed is increased. And the presence / absence of birds and beasts can be accurately identified.

The second divided region acquisition unit 23 divides the background difference image into x regions, and acquires x regions R _y (y is an integer of 1 to x). At this time, the size and shape of the region R _y are not particularly limited, but if the size and shape including the parked birds and beasts are set in advance, the presence or absence of the pigeon can be accurately identified. . x may be the same as the number m obtained by dividing the inter-frame difference image.

In the second divided region acquisition unit 23, for each of the regions R _y , when the number of pixels whose pixel value obtained by the difference process is equal to or greater than a preset threshold has a value equal to or greater than a preset threshold, It is preferable to perform a second inspection region selection process that is selected as the second inspection region R _yy to be processed by the stationary walking wildlife identification unit 24. When the number of pixels in the region _Ry is equal to or greater than a preset threshold value, there is a possibility that there is a bird or beast that is stopped or walking, so the region _Ry is selected as the second inspection region _Ryy . If the number of pixels in the region R _y is less than a preset threshold value, there is a low possibility that a bird or beast exists in the region R _y , and this is _regarded as a non-inspection region R _yn , and the processing is ended here. Although the second inspection area selection process is not essential, by performing this process, it is possible to remove an area where there is a low possibility of the existence of birds and beasts, so that the data processing speed thereafter can be increased.

The method of acquiring the second inspection region R _yy in the second divided region acquisition unit 23 is not limited to the above method, and a region having a predetermined size is scanned on the background difference image and arranged on the background difference image. All possible combinations are tried, and the number of pixels whose pixel values obtained by difference processing in each region are equal to or larger than a preset threshold value are checked, and the region R _y whose pixel number is equal to or larger than a preset threshold value _May be selected as the second examination region R _yy by determining that there is a possibility that there is a bird or animal that is stopped or walking.

The stationary walking wildlife identification unit 24 identifies the presence or absence of a stationary or walking wildlife for the region _Ry or the second inspection region _Ryy (hereinafter referred to as the second inspection region _Ryy ). By a known processing method that can be performed, it is determined whether or not the second inspection region R _yy is a second wildlife region where there is one bird that is stopped or walking. Examples of the processing method for identifying a stationary or walking bird and beast include a process using a color histogram, a process using template matching, or a process combining these.

In the process using the color histogram, first, a plurality of pieces of image data that are stopped in various postures are acquired in advance for the birds and beasts to be identified. For each of the acquired image data, a color histogram is created from a plurality of images by counting the frequency included in a specific division among a plurality of divisions obtained by dividing RGB values into predetermined numerical ranges. Then, normalization is performed and this is used as a reference color histogram. As an example, FIG. 10 shows a reference color histogram created for a stationary pigeon. For the second inspection region R _yy , a color histogram is created and normalized in the same manner, and this is used as the color histogram of the second inspection region. The similarity between the color histogram of the second inspection region and the reference color histogram is calculated using a value representing the distance between the histograms such as the Bhatterary distance. When the similarity with respect to the reference color histogram is equal to or less than a predetermined value, the second inspection region _Ryy is determined to be a second wildlife region where there is a stationary or walking wildlife.

In the process using template matching, first, a plurality of image data of parked birds and beasts to be identified are acquired in advance and used as template images. The second inspection region R _yy is scanned to take a pixel value difference from the template image for each pixel in the second inspection region R _yy and record the total number of pixels in which the difference is equal to or less than a predetermined value. . If the total number of recorded pixels is the predetermined value or more, as the template image as compared to the second inspection region R _yy is close image, the second inspection region R _yy exist wildlife or walking in retention It determines with it being the 2nd animal and beast presence area. When it is determined that the total number of recorded pixels is smaller than the predetermined value and there is no bird or animal that is stationary or walking in the second inspection region _Ryy , the same processing is performed using another template image. Repeat the process described above, when at least one of the plurality of template images is determined to be the second inspection region R _yy and near image, the second inspection region R _yy wildlife or walking in retention It is determined that this is the second wildlife region.

In the process using template matching, the template image and the second inspection region R _yy may be _converted from a color image into a gray image or a binary image in advance. By performing such processing, the subsequent data processing speed can be increased. FIG. 11 is an explanatory diagram showing an outline of a process for performing template matching by converting a template image and a second inspection region into a binary image. The area shown at the upper left in FIG. 11 is a template image. In the template image, the area where the birds and beasts are present is converted into a binary image in white, and the other areas are converted into black. The second inspection region R _yy is a binary image in which the region extracted by the difference processing is white and the other regions are black. On the second inspection region R _yy is scanned by the template image, the area of a region that overlaps with the white area in the template image in the second inspection region R _yy is a predetermined ratio with respect to the white area of the second inspection region R _yy When it is above, it determines with the 2nd inspection area _| region _Ryy being the 2nd bird and beast presence area | region in which the birds and beasts which are stopping or walking exist.

  In the process using both the color histogram and the template matching, for example, the determination by the color histogram is first performed as described above, and the template matching is further performed as described above for the region determined to be the second wildlife region. If it is determined that this region is also the second wildlife region, it is finally determined that it is the second wildlife region where there is one stationary or walking bird.

The wildlife identification correction unit 32 uses the determination result obtained by the flying wildlife identification process and the determination result obtained by the stationary walking wildlife identification process, so that each of the region R _sy and the region R _yy A re-determination process is performed to determine whether the region is a single wildlife region, the second wildlife region, or a wildlife region where none exists. The wild animal identification correction unit 32 is not essential, but by having the wild animal identification correction unit 32, the wild animal can be identified with higher accuracy.

The birds and beasts identification correction unit 32 can employ, for example, a re-determination process using a probability / statistical model (HMM: hidden Markov model) described below.
For example, the state of the birds and beasts on the image is the state S, the state of the birds and beasts in the image P _t-1 is the state S _i (t−1), the state of the birds and beasts in the image P _t is the state S _j (t), and the state S Let the probability of transition from _i to state S _j be the state transition probability a _ij .
Here, it is assumed that the following three states exist as bird and beast states in a specific region.
State S ₀ : State without birds and beasts State S ₁ : State where birds and beasts are parked or walking State S ₂ : State where birds and beasts flapping are present At this time, a subscript of the symbol “S” indicating the state I, j is 0, 1, or 2.
First, the state transition probabilities a _ij of all combinations that transition from each state S _i to each state S _j are set by acquiring data in advance. For example, the state transition probability a _ij is set as shown in FIG. In FIG. 12, for example, the state transition probability a ₀₁ for transition from the state S ₀ to the state S ₁ is set as a ₀₁ = 0.0225.

Next, it is assumed that only one symbol O _m (t) among the symbols shown below is generated in a specific region when the image P _t−1 transitions to the image P _t . That is, the determination results output from the above-described flying bird identification unit 14 and stationary walking bird identification unit 24 are indicated by the following symbols.
Symbol O _{0 (t):} the region R _st, fluttered by and symbol O ₁ when both wildlife wildlife and stationary or while walking is not present _(t): the region R _st, in the absence of wildlife that fluttered Symbol O ₂ (t): When there is a flapping bird in the region R _st , and when there is no stationary or walking symbol B ₃ (t): Region R When both stray birds and birds that are parked or walking are present in _st

Here, when a transition to each state S _j from each state S _i, considered output probability of the combination per symbol O _m of transitions for each state are different, the output probability b _ij of each symbol O _m ( O _m ) is set by acquiring data in advance. Note that m is 0, 1, 2, or 3 as a subscript of the symbol “O” indicating a symbol.

For example, the output probability of the symbol O ₀ when transitioning from the state S ₀ to the state S ₀ is b ₀₀ (O ₀ ) = 0.98 and the output probability b ₀₀ (O ₁ ) = 0. 01, output probability b ₀₀ (O ₂ ) = 0.01, output probability b ₀₀ (O ₃ ) = 0, output probability b ₀₁ (O ₀ ) = 0.4, output probability b ₀₁ (O ₁ ) = 0. 5... Can be set.

As described above, the state transition probability a _ij and the symbol output probability b _ij (O _m ) set in advance and the determination results actually obtained from, for example, the image P ₁ to the image P _t , that is, the sequence of the symbol O _m Is used to estimate the most probable state (maximum likelihood estimation) as the current state. As a method of maximum likelihood estimation of the state, for example, the following method of calculating only the final state can be adopted using the concept of the Viterbi algorithm.

Here, q is the number of the halfway image, t is the number of the final image for which maximum likelihood estimation is performed, U is the maximum value of the numerical value indicating the state, and U is 2 in this example.
Specifically, for the region R _s and region R _y in the image P _t , the state transition probability a _ij and the output probability b when the symbol O _m (1) is output as the determination result when q = 1. _The above (formula 1) and (formula 2) are calculated for all combinations of _ij (O _m (1)). Next, the state transition probability a _ij and the output probability b _ij (O _m when the symbol O _m (2) is output as the determination result when the state S (1) and q = 2 are obtained as a result of the calculation. (2) Calculate (Equation 1) and (Equation 2) above for all combinations. Such calculation is performed by incrementing q one by one from q = 1 to t, and the value of S (t) which is the final state is obtained, and the state S (t) at this time is represented by each region in the image P _t Let it be a result of the redetermination process of R _s and region R _y .

Note that the initial δ _i (0) is δ _i (0) = π _i . Here, π _i is the probability of the initial state i. π _i may be determined empirically from past data. For example, as shown in FIG. 8 (a), if there is no particular problem assuming that there is no bird and beast, _{Assume that 0} = 1 and π _j = 0 (j ≠ 0).

In the above example, three examples of the states S _{0 to} S ₂ have been described for the sake of simplicity. However, when it is determined that there is a bird and beast in a specific region in the image P _t , the number of states is changed. The same processing may be performed by adding a combination of the state of the specific region and the state of the region adjacent to the specific region as a new state.

In the above example, the case where only one type of wildlife is determined has been described for the sake of simplicity. However, in the case of determining multiple types of wildlife, the same processing may be performed by increasing the number of states. Good. Further, a determination result of the image P _t-r taken before the image P _t symbol O _{m (t-r),} for example, by utilizing the maximum likelihood estimation back now and at least the last 30 frames about The identification accuracy of birds and beasts can be improved.

The redetermination process in the wildlife identification correction unit 32 is not limited to the process based on the above-described probability / statistical model (HMM), and a method of improving the detection rate of the wildlife by the following redetermination process may be adopted. Good. For example, in the flying birds and animals identification unit 14, a region _{R s} in the image _{P t} for the wildlife absence regions not determined as the first wildlife existence area, in the region _{R s} in the image _{P 1} ~ image _{P t-1,} 1 One even if it was determined as the first wildlife existing region is likely to wildlife is present in this region R _s, it is likely to be the first wildlife presence region. Therefore, for example, when at least one first wildlife region is present in the region R _s in the image P _t-60 to the image P _t-1 going back 60 frames before the current image P _t. determines that the region R _s that has been determined that the first wildlife absence region is first wildlife presence region. Between the region R _s in the image P _{t-60 ~} image P _t-1, when the first wildlife existing area does not exist at all, the region R _s that has been determined that the first wildlife absence region Is determined to be correct, and is determined to be the first non-existing region. The number of frames going back is not particularly limited, and can be set as appropriate in consideration of the identification accuracy, the data processing speed, and the like. In addition, the criterion for re-determination that the region R _s that has been determined as the first wildlife-existing region is the first wildlife-existing region is not particularly limited to the above-mentioned criterion, and considering the identification accuracy, It can be set appropriately.

  In addition, after the re-determination process by the probability / statistical model (HMM), the re-determination process by the above-described method for improving the detection rate of the birds and beasts may be performed as post-processing, thereby increasing the detection power of the birds and beasts. Therefore, fail-safety can be ensured.

  The birds and beasts counting unit 33 counts the number of the first birds and beasts as the number of birds and beasts in a state of flapping, and the number of the second birds and beasts as the number of birds and animals that are parked or walking A stationary walking wildlife counting unit, a total wildlife counting unit that obtains the total number of birds and beasts that are flapping and the number of birds that are stationary or walking, and the first wildlife region and the first If there is an overlapping area with two birds and beasts, it is considered that they are double-counted, and from the total number of birds and birds, the number of birds that are flapping, or the number of birds that are parked or walking A bird count correction unit that reduces the count and obtains the number of birds and beasts that are flapping, the number of birds and animals that are stopped or walking, and the total number of birds and animals;

  The number of birds and animals that are flapping, the number of birds and animals that are stopped or walking, and the total number of birds and animals obtained by the birds and animals counting unit 33 may be displayed on the display device 8 or stored in the RAM 5 or the storage device 6. May be.

  Moreover, you may perform the emphasis process of enclosing the 1st birds and beasts presence area | region and the 2nd birds and beasts presence area | region which were determined to have birds and beasts on the display apparatus 8 with a frame line, etc., connecting these area | regions, A moving image may be created and stored, displayed, and transmitted to other devices.

  Next, the method for identifying the birds and beasts by the above-described bird and animal identification apparatus will be described with reference mainly to FIGS. FIG. 4 is a flow chart showing an embodiment of the method for identifying birds and beasts of the present invention.

  First, the imaging unit 2 such as a camera is installed at a predetermined location, and for example, a spatial region including a roof on which a pigeon may land is captured and acquired as a moving image, or is captured and captured at a predetermined time interval. Obtained as a picture (step S01). The captured image data is transmitted to the bird and animal identification device 1 and stored in the RAM 5 or the storage device 6. For image data, before or after being transmitted to the bird and animal identification apparatus 1, and before performing the processing described later, if necessary, a vertical and horizontal size reduction process and / or a compression process using a codec or the like is performed. If done, the amount of data can be reduced, and the data processing speed thereafter can be increased. As for a moving image, all continuously shot images may be stored, or a frame may be cut out as a still image at a predetermined time interval and stored.

Next, define a single image P _t to be processed, for detecting the birds and animals of the flight wildlife process and during retention in the roof and the ground or the like, or while walking for detecting birds and animals of the condition being fluttered before landing The stationary walking birds and beasts are processed individually. Note that the bird and animal identification device of the present invention only needs to be able to detect at least the birds and animals that are flapping before landing, so that the stationary walking and animal processing may be performed as necessary. First, the flying birds and beasts process will be described.

In the inter-frame difference processing unit 12, the image P _t and the image P _t-1 are read out, the difference process between the image P _t-1 and the image P _t is performed, an inter-frame difference image is created, and the RAM 5 or storage means 6 is temporarily stored (step S02). Although not specifically described in the following processing, the processed data is temporarily stored in the RAM 5 or the storage device 6 as necessary. From the inter-frame difference image, a region having a change between the image P _t and the image P _t−1 is extracted. For example, when the image P _t and the image P _t−1 are represented by RGB values for each pixel, and the difference value of the RGB value for each corresponding pixel is equal to or greater than a preset threshold value, the coordinates of the pixel change. It is saved as a certain area. As shown in FIG. 5 (a) and (b), when there is motion for pigeons are moved to the image P _t from the image P _t-1 to the position and orientation or the like of the pigeon between the image P _t The pigeon is extracted and shown as a white portion in the inter-frame difference image shown in FIG. The background without motion is not extracted and is shown as a black portion in the inter-frame difference image shown in FIG.

Next, the first divided region acquisition unit 13 performs division processing to divide the inter-frame difference image obtained by the difference processing into m regions, and m regions R _s (s is an integer of 1 to m). (Step S03). For example, in the inter-frame difference image shown in FIG. 6, the inter-frame difference image is divided into five vertical columns and five horizontal columns, and is divided into a total of 25 rectangular regions.

Next, if necessary, for each of the m regions R _s , when the number of pixels has a value equal to or greater than a preset threshold value, the first inspection region R _sy to proceed to the next flying bird identification process is selected. One inspection area selection process is performed (step S04). In the region R _s where the number of pixels is equal to or greater than the preset threshold value, there is a possibility that there is a flapping bird and beast, and this is set as the first inspection region R _{sy and the} process proceeds to the next flying bird and beast identification process. For the region R _s where the number of pixels is less than the preset threshold value, there is a low possibility that a bird or beast exists, so this is _regarded as the non-inspection region R _sn and the processing is ended here. By performing such a first inspection region selection process, m regions R _s can be narrowed down to the first inspection region R _sy where there is a possibility of the existence of birds and beasts, so that the data processing speed thereafter is increased. be able to.

Next, in the flying bird and animal identification unit 14, the region R _s or the first inspection region R _sy (hereinafter referred to as the first inspection region R _sy ) is assumed to be the body of the bird and beast and the wings of the bird and beast. It is divided into wing regions assumed as parts (step S05). As described above, as a method for dividing the first inspection region R _sy , various methods can be adopted as long as the method can be divided into the body and wing of the bird and beast. Here, a method of dividing the first inspection region R _sy into three regions adjacent in the horizontal direction will be described. As shown in FIG. 7, the first inspection region R _sy is divided into three regions adjacent in the horizontal direction, of which the central region R _syb is the body region and the regions R _sya and R _syc on both sides _thereof are the wing regions. Assume.

Next, for each of the body region and the wing region, the first inspection region R _sy is flapped by comparing the magnitude relationship between the number of pixels obtained by the difference processing and a preset threshold value. A flying wildlife identification process is performed to determine whether or not the first wildlife region is present (step S06). The region having a small number of pixels indicates that the region is poorly changed, and the region may be a birds and beast body. A region with a large number of pixels indicates a region that changes drastically, and the region may be a wing part of a bird or beast. When the number of pixels in the central region R _syb is smaller than a preset threshold value and the number of pixels in at least one of the regions R _sya and R _syc on both sides thereof is larger than the preset threshold value, the central region R _syb is defined as the body region. _Assuming that the regions R _sya and R _syc on both sides are the wing regions are correct, the first inspection region R _sy is determined to be the first wild region where there is one flapping bird. When the number of pixels in the central region R _syb is greater than or equal to a preset threshold value, or when the number of pixels in the regions R _sya and R _syc on both sides is less than or equal to a preset threshold value, the above assumption is denied, It is determined that there is no flapping bird in one inspection region R _sy .

  According to the bird and animal identification apparatus of the present invention, it is possible to detect the flapping birds and animals before landing by the processing so far, but when detecting the birds and animals that are stopped or walking together, Separate processing. The stationary walking wildlife processing will be described below.

In the background difference processing unit 22 reads the same image P _t as read during the flight wildlife process described above, a background image serving as a reference that has been saved by taking advance same spatial region as this, the image Difference processing between _Pt and the background image is performed to create a background difference image (step S12). In the background difference image, different areas are extracted between the image _Pt and the background image. For example, when the image P _t and the background image are represented by RGB values for each pixel, and the difference value of the RGB values for each corresponding pixel is equal to or greater than a preset threshold value, the coordinates of the pixels are the image P _t and the background It is saved as a different area from the image. For example, as shown in FIG. 8, as a result of the difference processing between the image P _t is a target of the difference processing shown in the background image and 8 shown in FIG. 8 (a) (b), in stationary roof or walking Is extracted, and is shown as a white portion in the background difference image shown in FIG. A region that is the same as the background image is not extracted, and is shown as a black portion in the background difference image shown in FIG.

Next, the second divided region acquisition unit 23 performs a dividing process of dividing the background difference image obtained by the difference processing into x regions, and x regions R _y (y is an integer of 1 to x). (Step S13). x may be the same as the number m obtained by dividing the inter-frame difference image.

Next, if necessary, for each of the x regions R _y , if the number of pixels has a value equal to or greater than a preset threshold value, the region is selected as the second inspection region R _yy that proceeds to the next stationary walking wildlife identification process A second inspection area selection process is performed (step S14). Regarding the region R _y having the number of pixels equal to or greater than a preset threshold value, there is a possibility that there is a bird or beast that is stationary or walking, and this is used as the second inspection region R _yy and proceeds to the next stationary walking and wildlife identification process. . For the region _Ry in which the number of pixels is less than the preset threshold value, there is a low possibility that a bird or beast exists, so this is set as a non-inspection region R _sn , and the processing is completed. By performing such a second inspection region selection process, x regions R _y can be narrowed down to the second inspection region R _yy where there is a possibility of the existence of birds and beasts, so that the data processing speed thereafter is increased. be able to.

Next, in the stationary walking wildlife identification unit 24, regarding the region _Ry or the second inspection region _Ryy (hereinafter referred to as the second inspection region _Ryy ), whether or not there is a wildlife that is stationary or walking A stationary walking wildlife identification process for determining whether or not the second inspection region _Ryy is a second wildlife existing region where there is one stationary or walking wildlife by a known processing method capable of identifying the Is performed (step S15). Examples of the processing method for identifying a stationary or walking bird and beast include a process using a color histogram, a process using template matching, or a process combining these.

In the following, processing using a color histogram will be described. A plurality of image data of stationary birds and beasts to be identified are taken in advance, a color histogram is created from the plurality of images as described above, normalized, and stored as a reference color histogram. Similarly, for the second inspection region R _yy , a color histogram is generated and normalized, and is temporarily stored as a color histogram of the second inspection region. The reference color histogram and the color histogram of the second inspection area are read and calculated using a value representing the distance between the histograms, such as the Bhatterary distance, and the similarity between the color histogram of the second inspection area and the reference color histogram is calculated. evaluate. When the calculated value is equal to or less than the predetermined value, the second inspection area R _yy is determined to be a second wildlife existence area where a stationary or walking wildlife exists. When the calculated value is larger than the predetermined value, it is determined that the second inspection region R _yy does not include a bird or animal that is stopping or walking.

  According to the bird and animal identification apparatus of the present invention, it is possible to individually detect the birds and animals that have fluttered before landing and the birds and animals that are stopped or walking by the processing so far, but further improve the identification accuracy of birds and animals. In such a case, the determination process is performed again by the bird / animal identification correction unit 32.

In the wildlife identification correcting unit 32, the determination result obtained by the flying wildlife identification process and the determination result obtained by the stationary walking wildlife identification process are corrected to correct the determination result up to the above, and the region R _sy and the region A re-determination process is performed to determine whether each of R _yy is the first wildlife region, the second wildlife region, or the non-existing wildlife region (step S21). . By performing this re-determination process, it is possible to more accurately identify the presence or absence of a bird or beast.

Below, the case where a re-determination process is performed using a probability and a statistical model (HMM) is demonstrated. First, as described above, the state transition probabilities a _ij of all combinations that transition from each state S _i to each state S _j are set and stored by acquiring data in advance. Further, the output probability b _ij (O _m ) of the symbol O _{m at} the time of transition from each state S _i to each state S _j is set and stored by acquiring data in advance.

The birds and beasts identification correction unit 32 reads out the state transition probability a _ij and the output probability b _ij (O _m ) and obtains them in the region R _sy and the region R _yy from the image P ₁ to the image P _t , for example. The most probable state is estimated as the current state using the determination result, that is, the symbol O _m series. As a method of maximum likelihood estimation of a state, for example, the concept of the Viterbi algorithm can be used. As an example of the state maximum likelihood estimation method, as described above, the transition state probability a _ij , the output probability b _ij (O _m ), the region R _sy and the region R _yy from the image P ₁ to the image P _t are obtained. The value of S (t) is obtained by using the series of symbols O _m and (Equation 1) and (Equation 2) described above, and this is used for re-determination processing of each region R _s and region R _{y in} the image P _t . As a result.

  Then, if necessary, the birds and beasts counting unit 33 counts the number of first birds and beasts as the number of birds and beasts in a state of flapping, and the number of the second birds and beasts is stopped or walked. A stationary walking wildlife counting process that counts as the number of birds in the middle, and a total wildlife counting process that obtains the total number of birds and birds that are flapping and the total number of birds and animals that are stationary or walking, If there is an overlapping area between the first wildlife area and the second wildlife area, it is regarded as double counting, and the total number of birds and the number of birds and birds that are flapping or being stopped Alternatively, the number of counts is subtracted from the number of birds and animals that are walking, and the number of birds and animals that are flapping, the number of birds and animals that are stopped or walking, and the number of birds and animals that are being walked and the total number of birds and animals are obtained (step S22). .

  Next, if necessary, the number of birds and animals that are flapping, the number of birds and animals that are stopped or walking, and the total number of birds and animals obtained by the birds and animals counting unit 33 are displayed on the display device 8 (step 8). S23).

In this way, and it ends the birds and animals discrimination processing for the image P _t. When performing wildlife identification processing for the next frame, the same processing as the image P _t for the next image P _{t + 1.}

  The bird and animal identification device, method and program thereof according to the present invention are used to prevent bird damage such as pigeons and crows, for example. In this case, a photographing unit such as a camera is installed in a place where it is necessary to prevent and avoid the invasion of pigeons and crows, and image data photographed by the photographing unit is transmitted to the bird and animal identification device, or a portable record When the image data is taken into the bird and animal identification device by the medium and the existence of the bird and animal is identified in the image, for example, a signal is transmitted to the animal and animal repellent device, and the bird and animal repellent device intrudes pigeons and crows, etc. Is prevented or driven away.

  A known device for preventing and avoiding the invasion of birds and beasts can be used as the birds and animals repelling device, for example, a device that irradiates light to birds and animals, a device that applies electric shock to birds and animals, and a device that sprays chemicals to birds and animals , A device that generates a odor that birds and beasts avoid, a device that generates sound for birds and animals, and the like.

  The wildlife identification apparatus of the present invention does not necessarily need to be installed so as to be close to the imaging unit and the wildlife repelling apparatus, and the wildlife identification apparatus, the imaging part, and the wildlife repellent apparatus are connected via the Internet or the like and can be remotely operated. It may be installed as follows.

  As described above, according to the wildlife identification apparatus, method, and program of this embodiment, it is possible to separately identify the wildlife that is flapping and the wildlife that is parked or walking, and the number thereof is displayed. Can be displayed on the device. In particular, since the flapping bird just before landing on the roof can be identified in real time, it is possible to take measures such as driving away the wild animal before or after landing. In addition, since the bird and animal identification device, method and program of this embodiment identify the birds and animals that are flapping, they invade into the space area where moving objects such as airplanes and birds in gliding are photographing. In this case, it is possible to suppress the possibility of misrecognizing it as a bird and beast, and to reliably identify the bird and beast to be identified. Furthermore, since the wildlife identification apparatus, method and program of this embodiment have the wildlife identification correction unit that re-determines using the past determination result, the wildlife can be identified with higher accuracy.

  The present invention is not particularly limited to the bird and animal identification apparatus, method and program thereof according to the above-described embodiment, and various modifications are possible within the scope of achieving the object of the present invention.

DESCRIPTION OF SYMBOLS 1 Bird and beast identification apparatus 2 Imaging part 3 CPU
4 ROM
5 RAM
6 storage device 61 program 62 data 7 input device 8 display device 9 processing unit 11 flying bird / beast processing unit 12 inter-frame difference processing unit 13 first divided area acquisition unit 14 flying bird / beast identification unit 21 stationary walking bird / beast processing unit 22 background difference processing unit 23 Second divided area acquisition unit 24 Stopped walking identification unit 31 Determination result processing unit 32 Wildlife identification correction unit 33 Wildlife counting unit

Claims (7)

For n images Pt (t is an integer of 1 to n) acquired by shooting a specific spatial region, an image Pt and an image Pt-r (r is 1 to 1) shot before the image Pt. an inter-frame difference processing unit that performs difference processing with each of at least one image Pt-r and obtains an inter-frame difference image;
A first divided region acquisition unit that divides the inter-frame difference image into m regions and acquires m regions Rs (s represents an integer of 1 to m);
The region Rs is divided into a body region that is assumed as the body of a bird and beast and a wing region that is assumed as a wing part of the bird and beast, and the size of the number of pixels obtained by the difference processing in each of the body region and the wing region is obtained. A flying wildlife identification unit that determines whether or not the region Rs is a first wildlife region where one fluttering wildlife exists,
A bird and animal identification device comprising:   The flying bird identification unit divides the region Rs into three regions that are adjacent in the horizontal direction, of which the central region is the body region and the regions on both sides thereof are the wing regions, and the number of pixels of the body region is When the number of pixels of the wing area is smaller than a preset value and the number of pixels of the wing area is greater than a preset value, the area Rs is determined to be a first wild beast existence area in which there is one fluttering bird. Item 2. The bird and animal identification device according to Item 1. A background difference processing unit that performs a difference process between the image Pt and a background image captured in advance for the specific space region, and obtains a background difference image;
A second divided region acquisition unit that divides the background difference image into x regions and acquires x regions Ry (y is an integer of 1 to x);
A stationary walking wildlife identification unit that determines whether or not the region Ry is a second wildlife existing region in which there is one stationary or walking wildlife;
The bird and animal identification device according to claim 1, comprising:   Using the determination result of the image Pt-r in the flying bird identification unit and / or the stationary walking bird identification unit, each of the region Rs and the region Ry is the first bird existence region, The birds and beasts identification apparatus of Claim 3 provided with the birds and beasts identification correction | amendment part which determines whether they are 2 birds and beasts presence area | regions, or the birds and beasts nonexistence area | region which none exists. A flying birds and beasts counting unit that counts the number of the first birds and beasts present area as the number of birds and beasts in a state of flapping;
A stationary walking wildlife counting unit that counts the number of the second wildlife existing area as the number of stationary or walking wildlife;
A total bird count unit for obtaining a total number of birds and beasts that is the sum of the number of birds and birds that are flapping and the number of birds and animals that are stopped or walking;
If there is an overlapping area between the first wildlife area and the second wildlife area, it is considered that the area is double-counted, and the total number of birds and the number of birds in the flapping state or the stop A bird count correction unit for subtracting the count from the number of birds in the middle or walking, and obtaining the number of birds in the flapping state, the number of birds in suspension or walking, and the total number of birds and animals,
The bird and animal identification device according to any one of claims 3 and 4. For n images Pt (t is an integer of 1 to n) acquired by shooting a specific spatial region, an image Pt and an image Pt-r (r is 1 to 1) shot before the image Pt. a difference process with each of at least one of the images Pt-r to obtain an inter-frame difference image;
Dividing the inter-frame difference image into m regions to obtain m regions Rs (s is an integer from 1 to m);
The region Rs is divided into a body region that is assumed as the body of a bird and beast and a wing region that is assumed as a wing part of the bird and beast, and the size of the number of pixels obtained by the difference processing in each of the body region and the wing region is obtained. And determining whether or not the region Rs is a first wildlife region where one fluttering wildlife exists;
A method for identifying birds and beasts. For n images Pt (t is an integer of 1 to n) acquired by shooting a specific spatial region, an image Pt and an image Pt-r (r is 1 to 1) shot before the image Pt. a difference process with each of at least one of the images Pt-r to obtain an inter-frame difference image;
Dividing the inter-frame difference image into m regions to obtain m regions Rs (s is an integer from 1 to m);
The region Rs is divided into a body region that is assumed as the body of a bird and beast and a wing region that is assumed as a wing part of the bird and beast, and the size of the number of pixels obtained by the difference processing in each of the body region and the wing region is obtained. And determining whether or not the region Rs is a first wildlife region where one fluttering wildlife exists;
A program that causes a computer to execute.

Images