JP3496563B2 - Object detection device, object detection method, and recording medium recording object detection program - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an object detection device for detecting an object in an image input from a surveillance camera or the like, and more particularly to an object detection device that is robust against illumination fluctuation.

[0002]

2. Description of the Related Art As a conventional method for detecting an object from a moving image taken by a surveillance camera or the like, there is known a method of distinguishing an object from a difference image between a background image created in advance and an image obtained at the present time. ing. As described in Japanese Patent Application Laid-Open No. 9-81714 for creating a background image, a feature value (for example, a brightness value) is calculated for each pixel for a plurality of images captured in the past, and a median of the plurality of feature values is calculated. Is updated as the pixel value of the background image of that pixel.

[0003]

However, in the method of creating the difference image between the background image and the current image, it is not robust against long-term illumination fluctuations, so that it cannot detect a moving object in a stable manner. There's a problem. The reason is that, when the illumination variation frequently occurs, the luminance value varies between the background image and the image captured at the current time due to the illumination variation.

The present invention has been made in view of the above circumstances, and provides an object detecting apparatus capable of detecting an object stably and highly accurately in response to frequent illumination fluctuations. The purpose is to provide.

[0005]

According to a first aspect of the present invention, there is provided an object detecting device for detecting an object using a picked-up image, wherein the object detecting device focuses on the picked-up image.
Luminance of the reference point set for each area and area of interest
Reference point selection means for extracting a value group from the image, and a region of interest
And the correlation of the brightness value between the reference point group and various lighting variations
Illumination fluctuation background model shown below and the reference point selection
A distance calculation means for calculating the distance between the obtained luminance value group from-option means, characterized in that a object determination means for determining the presence of an object based on the distance calculated in the distance calculation unit . The invention according to claim 2 is the object
The detection device includes a region of interest on the image and the reference point group.
Represents the correlation of brightness values under various lighting variations.
Illumination that creates a background model
In the changing background model generating means and the object discriminating means,
The background data corresponding to the pixel point determined to be the background
Using the brightness value group extracted by the reference point selection means
And a background creating means for updating
To do.

According to a third aspect of the present invention, the object detecting apparatus refers to the distance between the reference points from the size of the target object to be detected and the range in which the target object can exist when setting the reference point group. The present invention is further characterized by further comprising reference point setting assisting means for determining the position of the point, the arrangement direction of the reference point from the moving direction of the target object, and setting the reference point group.

According to a fourth aspect of the present invention, the object detecting device acquires a luminance distribution for each pixel from a background image group obtained in advance, and the illumination variation background model subjected to the illumination variation and the luminance distribution. It is characterized by further comprising reference point automatic setting means for selecting a pixel having a high degree of similarity and setting this pixel as a reference point.

According to a fifth aspect of the present invention, the distance calculation means creates an illumination variation model from a plurality of background data by using a statistical method, and the distance to the brightness value group obtained from the reference point selection means. Is calculated.

According to a sixth aspect of the present invention, there is provided an object detecting method for detecting an object using a picked-up image, wherein the object detecting method is a target area and a target area of the picked-up image.
Reference point selection processing for extracting the luminance value group of the reference point group set for each from the image, and the luminance of the attention area and the reference point group
The correlation of the values was expressed under various lighting fluctuations.
Lighting variation background model and obtained from the reference point selection process
A distance calculating process for calculating the distance between the luminance value group, and having an object determination process of determining the presence of an object based on the distance calculated in the distance calculation process.
The invention according to claim 7 is the object detection method,
The correlation of the brightness value between the attention area on the image and the reference point group
Illumination fluctuation background model represented under various lighting fluctuations.
Illumination fluctuation background model generation to create Dell from background data
Processing and the image judged as the background in the object discrimination processing
The background data corresponding to the raw point is used for the reference point selection processing.
Therefore, create a background that updates using the extracted luminance value group
And further processing.

According to an eighth aspect of the present invention, in the object detection method, when the reference point group is set, the distance between the reference points is referred from the size of the target object to be detected and the range in which the target object can exist. It is characterized by further comprising a reference point setting assisting process for determining the position of the point, the arrangement direction of the reference point from the moving direction of the target object, and setting the reference point group.

According to a ninth aspect of the present invention, in the object detecting method, a luminance distribution is acquired for each pixel from a background image group obtained in advance, and the illumination variation background model subjected to illumination variation and the luminance distribution. It is characterized by further including a reference point automatic setting process for selecting a pixel having a high degree of similarity and setting this pixel as a reference point.

According to a tenth aspect of the present invention, in the distance calculation processing, an illumination variation model is created from a plurality of background data by using a statistical method, and the distance to the luminance value group obtained from the reference point selection processing is calculated. Is calculated.

The invention described in claim 11 is a computer-readable recording medium in which an object detection program for detecting an object using a picked-up image is recorded, wherein the object detection program is focused on the picked-up image. Area and
Reference point selection processing for extracting a luminance value group of reference point groups set for each attention area from the image, attention area and reference point group
And the correlation of the brightness value under various lighting fluctuations
From the displayed illumination fluctuation background model and the reference point selection processing
Distance calculation processing for calculating the distance to the obtained brightness value group ,
It is characterized in that the computer is caused to perform an object discriminating process for discriminating the existence of an object based on the distance calculated in the distance calculating process. The invention according to claim 12 is
The object detection program includes a region of interest on the image and a previous region.
Correspondence of brightness value with reference point group
Is the illumination fluctuation background model shown below background data?
Illumination fluctuation background model generation process created by
The above-mentioned corresponding to the pixel point judged to be the background in another processing
Background data extracted by the reference point selection process
The background creation process that updates using the brightness value group
It is characterized by letting the computer do it.

In the thirteenth aspect of the present invention, the object detection program refers to the distance between the reference points from the size of the target object to be detected and the range in which the target object can exist when setting the reference point group. It is characterized in that the computer further executes a reference point setting assisting process for determining the position of the point, the arrangement direction of the reference point from the moving direction of the target object, and setting the reference point group.

According to a fourteenth aspect of the present invention, the object detection program obtains a luminance distribution for each pixel from a background image group obtained in advance, and the illumination variation background model subjected to the illumination variation and the luminance distribution. It is characterized in that a pixel having a high degree of similarity is selected, and the computer is further caused to perform a reference point automatic setting process for setting this pixel as a reference point.

According to a fifteenth aspect of the present invention, in the distance calculation processing, an illumination variation model is created from a plurality of background data by using a statistical method, and the distance to the brightness value group obtained from the reference point selection processing is calculated. Is calculated.

[0017]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An object detecting device according to an embodiment of the present invention will be described below with reference to the drawings. First, the principle of object detection in the same embodiment will be described. In the embodiment, it is assumed that the surface of the background object photographed by the image input means such as a video camera is a perfect diffuse reflection.

According to the document “Image Processing and Recognition” Yasuiin and Nagao (Shokoido), the brightness value I (x, y) at the point (x, y) on the obtained image is

[Equation 1] Where θ is the angle between the normal vector of the object surface and the ray vector, and ξ is a constant related to reflectance. Therefore, if the surface normal vector of the object and the ray vector of the light source are constant, the brightness value I (x, y) at the point (x, y) on the image is proportional to the constant relating to the reflectance.
However, if there is a change in illumination light,

[Equation 2] Becomes Here, Ψ (t) is illumination light that changes with time, and μ is an object surface reflection coefficient.

For example, the brightness values IA at two different points A and B on the same image, which are only influenced by the same light source,
IB is

[Equation 3]

[Equation 4] And on the IA and IB planes

[Equation 5] The relationship is established.

Therefore, when the brightness of the background pixel points changes under the influence of the same illumination change in a multidimensional brightness space having the brightness values of the background pixel points spatially separated on the same image as axes, The luminance distribution of these background pixel points is on the straight line shown by the equation (5) passing through the origin (luminance 0). The distribution of the brightness values of the background pixel points due to this illumination variation is called an illumination variation background model.

However, when an object having a surface normal vector and a surface reflection coefficient different from that of the background pixel enters these background pixels, brightness fluctuations due to disturbances other than illumination fluctuations occur, so that in a multidimensional space of brightness. Unlike the case of the illumination fluctuation, it deviates from the straight line (illumination fluctuation background model) of the equation (5).

For example, two points on the image shown in FIG.
In the cases of A and B, the distribution of the two points in the two-dimensional luminance space when the object does not exist at these two points and the illumination changes is shown in FIG.
A distribution that satisfies the illumination fluctuation background model (on the straight line of the equation (5)) is shown like the white dots in (c). Figure 5
In (c), the x-axis is the brightness value IA of the point A and the y-axis is the point B.
Is the brightness value IB.

On the other hand, as shown in FIG. 5B, when an object such as a person invades any one of the two points, the brightness value at the point where the object invades changes. The brightness distribution at this time is shown in FIG.
A distribution deviating from the illumination fluctuation background model (straight line of Expression (5)) is shown like a black dot in FIG.

Therefore, in a multidimensional luminance space having the luminance of spatially distant point groups on the image as an axis, the illumination fluctuation background model and the luminance distribution of spatially distant point groups on the image for object discrimination. It is possible to determine whether the point is the object or the background from the distance between and even if the illumination changes.

Next, a first embodiment of the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram showing the configuration of the first embodiment. In this figure, reference numeral 1
Is an image input unit that captures and holds an image output from a surveillance camera, a VTR, or the like, and the captured image may be a moving image or a still image. Further, the image input means 1 may be provided with a camera for taking an image. Reference numeral 2 is a reference point selecting means for selecting a reference point group set in advance for each pixel point of the image held in the image input means 1 and extracting a brightness value group of the reference point group from this image. Reference numeral 3 is a background creating means for updating the background data corresponding to the pixel point determined to be the background using the brightness value group extracted by the reference point selecting means 2.

Reference numeral 4 is obtained from the reference point selecting means 2 by creating an illumination variation background model in which the background changes under the influence of the illumination variation from the corresponding background data of the background creating means 3 for each pixel point of the image. It is a distance calculation means for calculating the distance between the brightness value group and the illumination fluctuation background model. Reference numeral 5 is an object discriminating means for discriminating a moving object from the distance calculated by the distance calculating means 4. Reference numeral 6 is an image output means for outputting the resulting image data output from the object discrimination means 5 to a display device or the like.

Next, the operation of object detection in the first embodiment will be described with reference to FIGS. FIG. 2 is a flowchart illustrating the operation of object detection. First,
The image input unit 1 reads an image captured by a camera or the like, and holds the brightness value for each pixel in a memory provided in the image input unit 1 (step S1). The image here is data of a two-dimensional array expressed by the brightness value of each pixel.

Next, for a certain pixel point P (x, y),
A reference point group having a different arrangement in the preset image space is acquired from the reference point selecting means 2, and the brightness value of the pixel corresponding to the reference point is read from the memory in the image input means 1 (step S2). . At this time, the reference point selection means 2
Selects a plurality of reference points having different arrangements on the image set in advance for each pixel point of the image, and extracts the brightness value of the reference point group from the image held in the image input means 1. As a method of setting the reference point group, nine points in a 3 × 3 area including the pixel point P (x, y) may be used, or in a simple case, adjacent points or pixel points separated by several pixels may be used. Good.

For example, the pixel point P (x,
3) and another pixel point different from this pixel as a reference point (that is, two reference points),
As shown in (a), when the reference point is set to the adjacent point in the x-axis direction, IA and IB in the equation (1) are I (x, y) and I (x +
1, y). Here, a case where two reference points are set will be described.

Next, the reference point selecting means 2 obtains the reference point group obtained in step S2 if the background data is not yet stored in the memory provided in the background creating means 3 for storing the background data. The brightness value group of is stored (step S
3). At this time, the background creating unit 3 selects the background data for each pixel point of the portion determined to be the background on the current image by the object identifying unit 5 to be described later, and the brightness value of the reference point group selected by the reference point selecting unit 2. Update with groups.

If the number of reference points for each pixel point on the image is M, the background data for each pixel point is given as an M-dimensional vector. Since the number of reference points is two in this embodiment,
The background data is a two-dimensional vector. When the background data is not created at the initial stage, the luminance value group for each pixel point of the reference point selection means 2 is used as the background data. If the first input image does not include a moving object, the background data basically need not be updated. However, due to the influence of noise or the like, the brightness value of the background area may vary due to factors other than the illumination variation, so the background data is updated at any time for the area regarded as the background by the object discrimination means 5 described later. Therefore, the error can be suppressed.

For example, as a method of updating when the reference points are two points PA and PB, a new addition is made at the pixel point P regarded as the background pixel by the object discriminating means 5 as shown in equation (6). Data can be weighted and updated gradually.

[Equation 6] Here, Pnew is new background two-dimensional data, and Pold is the background two-dimensional data so far. IA and IB are brightness values at points PA and PB, and α is a weighting constant.

Next, the distance calculating means 4 determines the target pixel point P.
The brightness value group of the reference point group corresponding to (x, y) is read from the memory in the background creating means 3. A distance D between the brightness value group of the reference point group and the straight line passing through the origin and the brightness value group of the current reference point group obtained in step S2 is calculated (step S4). At this time, the distance calculation unit 4 calculates the distance between the brightness value group for each pixel point extracted by the reference point selection unit 2 and the illumination fluctuation background model obtained from the background data generated by the background generation unit 3. . This calculation creates an illumination fluctuation background model by obtaining a straight line that passes through the background data of the background creating means 3 and the origin in a multidimensional space having the brightness value of the reference point group as an axis, and selects this straight line and the reference point. A method of obtaining the distance from the luminance value group for each pixel point extracted by the means 2 in this multidimensional luminance space is used.

For example, assume that two reference points are set for each pixel point on the image, and the background data of a pixel is (a,
b), if the brightness of the corresponding reference point group on the current image is (c, d), the point (a, b) on the two-dimensional brightness space
A distance D formed by a straight line passing through the point of origin and the point (c, d) is obtained by the equation (7).

[Equation 7]

Next, the distance D obtained in step S4 is compared with a preset threshold value T (step S5). If D> T, the pixel point P is determined.
It is determined that there is an object at (x, y), and if D = <T, it is determined to be the background (steps S6 and S7). At this time, the object discriminating means 5 discriminates whether or not it is an object from the distance D obtained by the distance calculating means 4. As a determination method, a pixel whose distance D is larger than a predetermined threshold value is regarded as an object region. By performing this discrimination processing for all pixels of the image (step S8), the object on the image at the current time can be detected. The detected object is output by the result image output means 6. When an image is displayed on the result image output means 6, if the display color of the pixel point P (x, y) determined to have an object is changed and displayed in step S5, only the object has a different display color. Is displayed,
The shape of the object can be confirmed.

If it is determined in step S8 that all pixel points have been processed, the number of pixel points determined to be an object is counted, and this count value is preset. It may be determined that an object exists when the count value is larger than the value. This makes it possible to detect only an object having a certain size.

On the other hand, the brightness value of the area determined not to be an object is output to the background creating means 3, and the background data is updated based on this brightness value (step S3).

Next, in step S8, when it is determined that the above-described determination processing is completed for all pixels, it is determined whether or not there is a next image (step S9).
If there is a next image, the process returns to step S1 and the above-described processing is repeated. When the processing is completed for all images, the object detection processing is completed.

By performing the above processing each time an image is captured, object detection can be performed sequentially.

Next, the second embodiment will be described in detail with reference to FIG. The embodiment shown in FIG. 3 is different from the embodiment shown in FIG. 1 in that a reference point setting assisting means 7 is further provided. In the above-described first embodiment, the user had to define the reference points in advance, but in the second embodiment, the reference point setting assisting means 7 for assisting the reference point setting performed by the user is newly provided.

The reference point setting assisting means 7 determines the arrangement of the reference point group according to the size and moving direction of the target object. When the surface reflection coefficient of the target object is equal to the background surface reflection coefficient, it is necessary to set the reference point groups apart from each other due to the size of the object. The reason is that, when an object with a low brightness value (black) covers the reference point group on the image, the brightness change of the reference point group may follow the illumination fluctuation background model with low brightness (near the origin). .

For this reason, the distance between the reference points is set in consideration of the size of the object to be detected on the image based on the information such as the camera position and direction given in advance. At that time, if the reference points are set apart from each other as shown in FIG. 5A, the range of the existing position of the object becomes wide even if the object can be detected. However, as shown in FIG. When the range is limited, it is possible to limit the existence range of the object by taking some of the reference points in a region where the object cannot exist.

Further, when the moving direction of the vehicle object is limited on the road as in the case of detecting the vehicle on the road, the reference point is arranged along the moving direction of the object to detect the moving object. It is possible to increase the sensitivity and reduce noise other than the moving direction.

For example, assuming that the moving object changes its brightness value per unit distance by T with respect to the moving direction, the change amount V of the brightness value at the reference point is the vector of the moving direction of the moving object and the arrangement of the reference points. It is represented by the formula (8) by the angle formed by.

[Equation 8] Here, V is the change amount of the brightness value of the reference point, T is a constant, and φ
Is the angle between the traveling direction vector and the reference point sequence vector (0
(° ≦ φ ≦ 90 °), and W is the object moving distance.

In order to increase the detection sensitivity of the object, the change in the brightness of the reference point where the object exists may be increased. Therefore, in order to increase the sensitivity, V may be maximized, and φ needs to be 0 ° when T and W are constant.
That is, it is preferable to arrange the reference points along the moving direction of the object in order to reduce the angle formed between the traveling direction vector and the reference point group column vector. In the example of FIG. 7, the angle between the reference point group sequence vector in the reference point group sequence 1 and the moving direction of the object is φ1, and the angle between the reference point group sequence vector in the reference point group sequence 2 and the moving direction of the object is If φ2, then from φ1 <φ2,

[Equation 9]

[Equation 10]

[Equation 11] Becomes

Compared to the reference point group sequence 2, the reference point group sequence 1 has the reference points arranged along the moving direction of the object, so that the luminance change amount of the object at the reference point increases. Therefore, the reference point group sequence 1 is more sensitive to luminance change than the reference point group sequence 2. On the contrary, regarding the amount of change in luminance with respect to the variation perpendicular to the moving direction, the reference point group sequence 1 is less sensitive to the luminance change than the reference point group sequence 2 as shown in equations (12) to (14). That is, it becomes insensitive to the moving direction deviated from the reference point group column vector, and has the effect of lowering the sensitivity to noise.

[Equation 12]

[Equation 13]

[Equation 14]

Next, the third embodiment will be described in detail with reference to FIG. The embodiment shown in FIG. 4 differs from the second embodiment shown in FIG. 3 in that a reference point automatic setting means 8 is provided instead of the reference point setting assisting means 7. In the first and second embodiments described above, the user had to set the reference points in advance, but in the third embodiment, the reference point automatic setting means 8 for automatically setting the reference points is provided. It was

The reference point automatic setting means 8 automatically sets the reference point group for each pixel point used in the reference point selecting means 2. The reference point group is set as follows. It is assumed that M background images including only the background are captured in advance, and each of them is the background images 1 to M.
To call. The brightness value of the pixel point P on the background image i (1 ≦ i ≦ M) is Pi, and the brightness value of the pixel point P ′ of the reference point candidate with respect to the pixel point P is Pi ′.

At this time, the variance-covariance matrix S is expressed by equations (15)-
Given in (17).

[Equation 15]

[Equation 16]

[Equation 17] Here, the minimum eigenvalue λ2 of the eigenvalues λ1 and λ2 (λ1> λ2) of the covariance matrix S given by the equations (15) to (17) is set as the score of the reference point candidate P ′ with respect to the pixel point P. λ2 indicates the degree of variation of the pixel point P and the reference point candidate pixel point P ′ with respect to the illumination fluctuation background model when the luminance changes. For this reason, reference point candidates for the pixel point P are taken for all pixel points on the image, and λ2 for each reference point candidate.
Is calculated and set as reference points in the order of reference point candidates having smaller λ2. The number of reference points may be set to a preset number, or a score smaller than a preset value (λ
You may set only the reference points with 2).

In the distance calculation means 4 of the above-mentioned first, second and third embodiments, to determine whether or not it is an object,
Although the distance D between the point and the straight line (given by the equation (7) in the two-dimensional example) is used, in addition, the illumination fluctuation background model, the brightness value of the reference point group of the current image, and the illumination fluctuation background model are used. It is also possible to use the angle formed by a straight line. further,
Even when the surface of the background object does not satisfy the assumption of perfect diffuse reflection, that is, when the illumination fluctuation background model does not match the straight line passing through the origin, the distance calculation means 4 can calculate the distance by using the statistical analysis. When using statistical analysis, it is necessary to store multiple background data at different times.

In the following description, the number of reference points is N and the number of background data used is M. First, the case of using the principal component analysis will be described. Reference point P corresponding to pixel point P
Background data j (1 ≦ j ≦ M) at i (1 ≦ i ≦ N)
Let Pij be the luminance value at, and the variance-covariance matrix S of N-dimensional variate data with N reference point luminance values as variables is
It is represented by 8) to (20).

[Equation 18] However,

[Formula 19]

[Equation 20] Is.

N of the variance-covariance matrix S in equation (18)
Let eigenvalues be λ1, λ2, ..., λN (λ1 ≧ λ2 ≧ ... ≧ λN), and let the eigenvector corresponding to the maximum eigenvalue λ1 be V1. The straight line representing the first principal component is the average value point of each reference point brightness value.

[Equation 21] And becomes a straight line of the direction vector V1. In the distance calculation of the pixel points of the current image for object detection, the straight line of the first principal component is used as the illumination fluctuation background model, and the luminance value (Pc1, ..., PcN) of the reference point group of the pixel points Pc of the current image and the This is performed by obtaining the distance from the straight line of one principal component.

Next, the case where the Mahalanobis distance is used will be described below. Similar to the above-described principal component analysis, N reference points corresponding to each pixel point are used as variables, and a variance-covariance matrix S of N variable data having M background data as samples is obtained from Equation (18), Average of reference points of background data

[Equation 22] Then, the Mahalanobis distance D2 is calculated from the luminance values (Pc1, ..., PcN) of the reference point group corresponding to the pixel points of the current image for which the object detection is performed, based on the equation (21).

[Equation 23]

Details of principal component analysis and Mahalanobis distance are described in "Practice of Multivariate Analysis (1)" by Kanro Kan (Hyundai Mathematics Co., Ltd.) and "History of Multivariate Analysis" by Arima and Ishimura (Tokyo Book).

In addition, each function shown in FIGS.
The object detection process may be performed by recording a program for realizing each step shown in (1) in a computer-readable recording medium, and causing the computer system to read and execute the program recorded in the recording medium. The “computer system” mentioned here includes an OS and hardware such as peripheral devices. The "computer-readable recording medium" means a portable medium such as a floppy (registered trademark) disk, a magneto-optical disk, a ROM, a CD-ROM, or a storage device such as a hard disk built in a computer system. .

Further, the "computer-readable recording medium" means a program dynamically for a short time like a communication line for transmitting the program through a network such as the Internet or a communication line such as a telephone line. The one that holds the program, and the one that holds the program for a certain period of time, such as the volatile memory inside the computer system that serves as the server or the client in that case, are also included. Further, the above program may be one for realizing some of the functions described above, and may be one that can realize the above functions in combination with a program already recorded in the computer system.

[0057]

As described above, the first and second aspects of the present invention are as follows .
According to the inventions described in 6 , 7 , 11, and 12, it is possible to obtain an effect that stable person detection can be performed even when there is a rapid change in illumination. The reason is that by obtaining the multidimensional luminance distributions of spatially different reference points, the rapid luminance variation is limited to the linear variation in the luminance space like the white dots in FIG. The above effect is not taken into account, and only slight deviations in the direction perpendicular to the straight line caused by fluctuations in the light source position due to the movement of the sun or the effects of noise are corrected, and the background and object Since it is determined whether or not the person is detected, stable person detection can be performed.

Further, according to the invention described in claims 3, 8 and 13 , it is possible to obtain the effect of being robust against noise other than the target object by selecting the reference point matched to the detection target. The reason is that it is possible to give a filter effect regarding the size and the direction depending on how the reference points are arranged.

According to the invention described in claims 4, 9 and 14 , the effect that the user does not have to set the reference point in advance can be obtained. The reason is that a stable set of reference points is selected from the background image in advance and the reference points are set.

According to the invention described in claims 5 , 10 and 15 , there is an effect that the object can be detected even when the assumption of perfect diffuse reflection is not satisfied on the surface of the background object. The reason is that an illumination variation background model is created from a plurality of background data using a statistical analysis method, and the distance is calculated.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a first exemplary embodiment of the present invention.

FIG. 2 is a flowchart for explaining the operation of the first embodiment.

FIG. 3 is a block diagram showing a configuration of a second exemplary embodiment of the present invention.

FIG. 4 is a block diagram showing a configuration of a third exemplary embodiment of the present invention.

FIG. 5 is a diagram for explaining a specific example of the operation of the first exemplary embodiment.

FIG. 6 is a diagram showing a specific example of the operation of the second exemplary embodiment.

FIG. 7 is a diagram showing a specific example of the operation of the second exemplary embodiment.

[Explanation of symbols]

1 ... Image input means, 2 ... Reference point selection means, 3 ... Background creation means 4 ... Distance calculation means, 5: object discrimination means, 6 ... Image output means, 7 ... Reference point setting aid, 8 ... Automatic reference point setting means.

Continuation of the front page (56) Reference JP-A-5-35874 (JP, A) JP-A-7-302328 (JP, A) JP-A-8-221577 (JP, A) JP-A-6-52311 (JP , A) Japanese Patent Laid-Open No. 2000-348163 (JP, A) Vehicle extraction method using spatial differentiation and difference processing, IEICE Transactions, Japan, November 1997, Vol. J80-D-I I No. 11, pp. 2976-2985 Background subtraction method robust against changes in illumination, Information Processing Society of Japan, Research Report, Japan, March 19, 1999, Vol. 99 No. 29, pp. 17-24 (58) Fields investigated (Int.Cl. ⁷ , DB name) G06T ⁷ /00-7/60 G06T 1/00

Claims (15)

(57) [Claims] 1. An object detection device for detecting an object using a captured image, wherein the object detection device is set for each attention area of the captured image and each attention area.
The reference point selecting means for extracting the brightness value group of the reference point group from the image and the correlation between the brightness values of the attention area and the reference point group are various.
Lighting fluctuation background model expressed under lighting fluctuation and
Wherein a distance calculating means for calculating a distance between the obtained luminance value group from the reference point selection means, that a object determination means for determining the presence of an object based on the distance calculated in the distance calculation unit Object detection device. 2. The object detecting device is configured to correlate luminance values between a region of interest on the image and the reference point group.
Illumination variation showing relationships under various illumination variations
Create a background model from background data Lighting background model
And a pixel point determined to be the background by the object determining means.
The corresponding background data is acquired by the reference point selecting means.
Ri the issued and background generation means for updating using the luminance value group
The object inspection according to claim 1, further comprising:
Output device. 3. The object detecting apparatus, when setting a reference point group, the distance between the reference points from the size of the target object to be detected, the position of the reference point from the possible existence range of the target object, the target object 2. The object detection apparatus according to claim 1, further comprising reference point setting assisting means for respectively determining the arrangement direction of the reference points from the moving direction of the reference point and setting the reference point group. 4. The object detection device acquires a luminance distribution for each pixel from a background image group obtained in advance, and selects a pixel having a high degree of similarity between the illumination fluctuation background model subjected to illumination fluctuation and the luminance distribution. The object detecting apparatus according to claim 1, further comprising a reference point automatic setting unit that sets the pixel as a reference point. 5. The distance calculation means creates an illumination variation background model from a plurality of background data by using a statistical method, and calculates the distance from the brightness value group obtained from the reference point selection means. The calculation is performed according to claim 1.
4. The object detection device according to any one of 4 above. 6. An object detection method for detecting an object using a captured image, wherein the object detection method is set for each attention area of the captured image and each attention area.
The reference point selection process for extracting the brightness value group of the reference point group from the image and the correlation between the brightness values of the attention area and the reference point group are various.
Lighting fluctuation background model expressed under lighting fluctuation and
A distance calculating process for calculating the distance between the obtained luminance value group from the reference point selection process, and characterized by having the object determination process for determining the presence of an object based on the distance calculated in the distance calculation process Object detection method. 7. The object detection method is a correlation of luminance values between a region of interest on the image and the reference point group.
Illumination variation showing relationships under various illumination variations
Create a background model from background data Lighting background model
The pixel generation process and the pixel point determined to be the background in the object discrimination process.
The corresponding background data is acquired by the reference point selection processing.
Ri the issued and background creation process of updating by using the luminance value group
The object detection according to claim 6, further comprising:
Method. 8. The object detection method, when setting the reference point group, the distance between the reference points from the size of the target object to be detected, the position of the reference point from the possible existence range of the target object, the target object 7. The object detection method according to claim 6 , further comprising a reference point setting assisting process for determining a placement direction of the reference points from the moving direction and setting the reference point group. 9. The object detecting method obtains a luminance distribution for each pixel from a background image group obtained in advance, and selects a pixel having a high degree of similarity between the illumination fluctuation background model subjected to illumination fluctuation and the luminance distribution. The object detection method according to claim 6 , further comprising a reference point automatic setting process for setting this pixel as a reference point. 10. The distance calculation process creates an illumination variation model from a plurality of background data using a statistical method, and calculates a distance from a brightness value group obtained from the reference point selection process. The object detection method according to any one of claims 6 to 9 . 11. A computer-readable recording medium in which an object detection program for detecting an object using a captured image is recorded, wherein the object detection program is set for each attention area of the captured image and each attention area. Been
The reference point selection process for extracting the brightness value group of the reference point group from the image and the correlation between the brightness values of the attention area and the reference point group are various.
Lighting fluctuation background model expressed under lighting fluctuation and
A distance calculating process for calculating the distance between the obtained luminance value group from the reference point selection process, to carry out the object determination process in a computer for determining the presence of an object based on the distance calculated in the distance calculation process A recording medium having an object detection program recorded therein. 12. The object detection program is configured to correlate luminance values between a region of interest on the image and the reference point group.
Illumination variation showing relationships under various illumination variations
Create a background model from background data Lighting background model
The pixel generation process and the pixel point determined to be the background in the object discrimination process.
The corresponding background data is acquired by the reference point selection processing.
Ri the issued and background creation process of updating by using the luminance value group
The method further comprising a computer
Recording medium recording the object detection program according to 11. 13. The object detection program, when setting a reference point group, determines the distance between the reference points from the size of the target object to be detected, the position of the reference point from the possible existence range of the target object, and the target object. computer arrangement direction of the reference point from the direction of movement was recorded object detection program according to claim 11, characterized in that causing the computer reference point setting assistance process to set a reference point group respectively determined to further the A readable recording medium. 14. The object detection program acquires a luminance distribution for each pixel from a background image group obtained in advance, and selects a pixel having a high degree of similarity between the illumination fluctuation background model subjected to illumination fluctuation and the luminance distribution. and, further con reference points automatic setting processing for setting the pixel as a reference point
Computer readable recording medium recording the object detection program according to claim 11, characterized in that to perform the computer. 15. The distance calculation processing is characterized in that an illumination variation model is created from a plurality of background data by using a statistical method, and a distance to a brightness value group obtained from the reference point selection processing is calculated. Claims 11 to 1
A computer-readable recording medium in which the object detection program according to any one of 4 is recorded.