JP3047658B2 - Method for determining binarization threshold in moving object detection - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for determining a binarized threshold value in moving object detection using image processing.

[0002]

2. Description of the Related Art In general, in a moving object detection using image processing, a method of detecting a moving object includes an inter-frame difference method for obtaining a difference between frames, and a corresponding point for calculating a corresponding point between images of continuous images. And a background subtraction method for taking the difference between the background image and the input image.

The inter-frame difference method is a method of calculating a difference between two images input at a certain time interval. Therefore, although the configuration is simple, the region obtained as a result of the difference does not represent the symmetrical object itself, and hence it is difficult to extract the image of the moving object from only the result.

In the corresponding point method, an image is divided into small areas.
This is a method of calculating a shape parameter characterizing each small area and calculating a corresponding point by correlating the shape parameter between images. For this reason, when the processing is heavy and the rotation or shape of the object changes, the shape parameter of each small area changes, and it becomes difficult to detect the corresponding point.

The background subtraction method is a method of obtaining an image of an object to be extracted by subtracting a background image from an input image including the object to be extracted and the background image. This method
If there is a correct background image, the target object can be satisfactorily extracted. Therefore, it is necessary to capture a background image that does not include the extraction target object in advance. However, if the background changes, a correct result cannot be obtained. Therefore, the method of generating the background becomes a problem.

[0006]

In the background subtraction method in the conventional motion detection, the threshold value for binarization and the background integration parameter are individually set. The threshold value for binarization is set uniformly for the image,
There is a method in which an image is divided into regions according to the difference in the reflectance of the background and set for each region. As for the background integration parameter, there has been a method of measuring a background change amount and adaptively setting the change amount for each change region. Among them, a method for setting uniform integration parameters for a motion extraction processing area in an image is disclosed in Japanese Patent Application Laid-Open No. Sho 62-1140.
There is a method described in JP-A-64. This method is based on the assumption that an input image f (n) and an old background image are g (n-1).
The background image g (n) to be updated is g (n) = A × f (n) + (1−A) × g (n−1) (1). Here, A is a constant and is common to all pixels. However, according to this method, there is a problem that a hole or a tail is generated in the region to be extracted due to the moving speed of the extraction target object, the reflectance of the surface, and the threshold, and the shape of the extraction region is deformed.

SUMMARY OF THE INVENTION The present invention solves such a conventional problem. In the present invention, there is provided a moving object detection method capable of improving the accuracy of an extraction area of an extracted moving object when a background is updated using a background integration method. It is an object of the present invention to provide a method of determining a threshold value.

[0008]

According to the present invention, in order to achieve the above object, a difference value image between an input image and a background image is obtained.
The threshold value is set to half of the difference value image for each pixel.

According to the present invention, a threshold value for binarization and a background integration parameter are determined by using a difference value image between an input image and a background image and an estimated value of a time at which the object to be extracted stops moving. Things.

[0010]

According to the present invention, accurate motion extraction can be performed by the above method. In addition, when uniform integration parameters are set for a motion extraction processing area in an image,
An estimated value of the time at which the extraction target object stops moving from the input image is calculated, and a threshold value is determined based on a difference value between the calculation result and the background of the target object having a change due to the reflectance of the surface of the extraction target object. By determining the background integration parameters, it is possible to extract a target object with high accuracy.

[0011]

DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a flow diagram showing a process for performing a moving object extraction process using a background image according to an embodiment of the present invention.

As shown in FIG. 1, an image is first input in step 100, and then an input image I is input in step 101.
(I, j) is thinned out by m in the i direction and n is thinned out in the j direction to create a processed image. Here, a block of m × n pixels is defined as one processing pixel, and its level is an average value of each pixel. By taking the average value of each pixel in this way, it is possible to reduce fixed noise of each pixel and noise due to dark current.

Further, the timing for obtaining the difference between the input image and the background image is determined by the number of frames. For example, when five frames are used, the frame thinning rate is expressed as 5. This process is also performed by the thinning process of step 101. The reason why the frame thinning rate is variable is that the moving speed of the extraction target object is not always constant, and it is necessary to change the processing according to the moving speed. Details will be described in the description of the threshold control processing step 106.

The processed image f thinned out in step 101
(N) and the background pixel g (n-1) set in the background memory by the function setting processing step 104 and the integration control processing step 103 by the CPU, in step 102
The new background pixel g (n) is updated and stored by the following equation.

G (n) = a × f (n) + (1−a) × g (n−1) (2) The processed image thinned out in step 101 and step 102
And the background pixel updated in step 11
Based on the threshold value obtained in step 106, which is controlled based on the information obtained in step 6, a difference process is performed in step 105 to obtain a binarized image.

Here, the value of the integration parameter a in the above equation (2) is uniform for the input image. As a method of updating the background, there is a method of creating the background without updating the background of the area where the motion is extracted. However, it is necessary to make the value of the integration parameter a uniform with respect to the input image for the following reasons. did. (1) The method of generating the background is simplified, the burden on hardware and software is reduced, and the cost is reduced. (2) By simplifying the method of generating the background, if the background has a non-regularity such as in the field of monitoring, random tuning is omitted and maintenance-free is realized.

However, if the value of the integration parameter a is made uniform with respect to the input image, the area of the object to be extracted is also updated as a background, and the time to stop moving the object, the threshold for binarization, Due to the difference in the reflectance of the surface, as shown in FIG.
And the like, and the background image is extracted as a motion in an area having a different reflectance due to a change in the light source. In the former, the difference in the reflectance of the surface of the object is set for each extraction target object from the difference between the input image and the background image, and the time to cut off the movement of the object is estimated and calculated. By determining the background integration parameter from the threshold value of the binarization and the estimated time at which the object stops moving, the occurrence of holes and tails can be prevented. As for the latter, the estimation calculation of the extraction area and the determination of the respective thresholds are performed in the same manner as the former, based on the extracted result, and the size of the extraction area is set to the size of the extraction target object ( Pixels) range,
It is possible to cope with the removal by the large / small area removal processing in step 111 and the restriction on the variation in the size of the extraction area, and when the limit value is exceeded, by removing the area.

Here, the difference in the reflectance of the surface of the former object is set for each of the objects to be extracted from the difference between the input image and the background image, and the time at which the object stops moving is estimated and calculated. A method of determining a background integration parameter from a threshold for binarization and an estimated time at which the object will stop moving will be described.

First, the time at which the object stops moving is extracted.
The moving direction of the area can be obtained from the trajectory memory by the previous process.
The current center of gravity and the currently processed center of gravity are obtained. Then, each
The moving direction, the moving speed,
Measure the number of pixels in the extraction area in the moving direction and stop moving the object
Estimate the time. Next, the background image F (i, j) is entered.
Create a difference image C (i, j) of the force image I (i, j),
Allocate C (i, j) / 2 to each pixel (i, j)
Spatial threshold C_thmake. C created _th(I,
The difference image is binarized using j). Obtained here
C (i, j) and C_thFrom (i, j), the following two equations are
To create an optimal background integration parameter a.

[0020]

(Equation 1)

[0021]

(Equation 2)

Here, t is the time at which the object stops moving, Equation (3) is the range of the background integration parameter a where no hole is generated, and Equation (4) is the background integration parameter a where no tail is generated.
Range. FIG. 3 is a graph of this range. The relationship between t ₀ , t ₁ , and t ₂ is t ₀ > t ₁ > t ₂ . FIG. 4 is a graph extracted from the graph when t = t ₀ . In FIG. 4, the value of the integration parameter a in which neither a hole nor a tail occurs when t = t ₀ is in the range of the hatched portion. What is necessary is just to determine the value of the integration parameter a so as to fall within the shaded portion, and the background can be updated using this integration parameter a.

Further, a lower limit value is set for the threshold value in order to remove a noise generated when an input image is obtained due to a fixed noise or the like, with respect to the spatial threshold value C _th (i, j). There is a need. When obtaining the difference value C (i, j) between the input image and the background image, the lower limit value may be determined based on the result of measuring the temporal variation of each pixel.

In addition, since the range of t changes dynamically depending on the moving speed of the object, the range of the hatched portion in FIG. 4 greatly changes. As a countermeasure, as described above, a range of the moving speed of the object may be set in advance, and a frame thinning rate suitable for the range may be selected. For example, a method of providing a switch that can adjust the range of the moving speed of the object in three stages, a method of estimating the time to stop moving the object as in this case, and determining the value from the result of statistical processing are used.

At step 107, an M × M median filter is applied to the binary image obtained at the difference processing step 105 at M / 2. This is effective in two points, that is, noise can be removed and the size of the surface can be absorbed by adjusting the size. Of the latter, the size of the target object on the processed image is determined from the distance between the camera and the object, the actual size of the target object, the spatial thinning rate, the focal length of the lens, the number of pixels of the CCD, and the aspect ratio. Therefore, if each piece of information is obtained, the size of the median filter may be determined based on the size of the extraction target object.

The result of the median filtering in step 107 is masked in step 108 to limit the processing area. A mask is created in step 109 using the mask pattern. Next, labeling processing is performed on the processed image obtained as a result of the mask processing in step 110, and the process proceeds to the next step 111.

In step 111, area connection processing and large / small area removal are performed. Here, the size of the extraction area is compared with the range of the size of the extraction target object input in advance, and processing for removing an object that does not fit in the range is performed.
This makes it possible to remove noise components, such as snow, dogs, and cats, which are clearly different in size from the extraction target object, in the input image. Further, the image and the trajectory subjected to the area connection and the large / small area removal processing are subjected to a situation judgment (calculation of an estimated value of the time to stop moving) from the processing result of step 113, and the threshold control processing is performed in step 106 using the result. And update the control information. In the image from which the region connection processing and the large and small regions have been removed in step 111, the respective centers of gravity are set in step 11
2, the locus of the barycenter of each label obtained as a result is obtained in step 113, the locus is stored in the locus memory in step 114, and the display is switched in step 115.

[0028]

As described above, according to the present invention, the difference value image D between the input image I (i, j) and the background image F (i, j) is obtained.
By obtaining (i, j) and setting the threshold to D (i, j) / 2, the extraction of the extraction target object can be satisfactorily performed without deformation of the extraction area. When uniform integration parameters are set for a motion extraction processing area in an image, the input image I (i, j) and the background image F (i, j)
By using the difference value image D (i, j) and the time to cut off the movement of the extraction target object, the binarization threshold value and the background integration parameter are determined. In addition, it is possible to extract a motion that is maintenance-free and has a simple configuration without changing the method of updating the background.

[Brief description of the drawings]

FIG. 1 is a flow diagram showing background integration processing according to an embodiment of the present invention.

FIG. 2 is a schematic diagram showing an example of occurrence of distortion in an extraction region caused by uniformly giving a background integration parameter a to a screen in the embodiment.

FIG. 3 is a graph showing grounds for setting a threshold value to の of a difference value between an input image and a background image in the embodiment.

FIG. 4 is a graph showing a region where holes and tails do not occur when t = t ₀ in the embodiment.

[Explanation of symbols]

 120 Extraction target area 121 Hole 122 Tail 123 Extraction result

Claims (2)

(57) [Claims] 1. An input image I (i, j) and a background image F
A method of determining a binarized threshold value in moving object detection, characterized in that a difference value image D (i, j) of (i, j) is obtained and a threshold value is set to D (i, j) / 2. 2. When a uniform integration parameter is set for a motion extraction processing region in an image, an input image I (i,
j) and the difference value image D (i, j) between the background image F (i, j)
A threshold value for binarization and a background integration parameter using the time at which the extraction target object stops moving.