JPH10105689A - Moving body movement detector - Google Patents

Abstract

PROBLEM TO BE SOLVED: To detect the movement of an individual vehicle with a high precision without error. SOLUTION: The difference value between a background picture frame 15 previously photographed without vehicles and a preceding picture frame 14 photographed just before is binarized, and this binarized difference picture 16 is masked with individual vehicle area mask pictures 17 and 18 obtained from a vehicle area mask picture generation part 10, and these mask areas are cut out to generate templates. Since this template of each vehicle obtained by cutting out has the noise eliminated and is accurate, the movement vector of each vehicle is detected with a high precision by matching to a current picture frame.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a moving body behavior detecting device for photographing a moving body such as a vehicle using a camera and examining its behavior.

[0002]

2. Description of the Related Art Next-generation road traffic systems have been actively studied for the purpose of increasing road traffic capacity, suppressing traffic accidents, and saving energy. In such a system, a technology for photographing a vehicle traveling on a road and detecting the behavior of each vehicle is required. The following technologies have been introduced for this purpose (Literature: Takeshi Yasuiin, Shota Tsukanaka, Hiroshi Nagahashi: Extraction of passing vehicles by image processing:
IEEJ Spring Spring, D-627 (1994)). In this type of system, a road where a vehicle is not shown is photographed in advance, and a vehicle traveling on the road is repeatedly photographed at predetermined time intervals. Then, a motion vector for each vehicle is obtained by comparing the captured images.

[0003]

The above-described conventional vehicle behavior detecting device has the following problems to be solved. In a conventional system, it is determined whether or not a vehicle exists on a road by calculating a difference value between an image obtained by capturing only the road and an image obtained by capturing a vehicle traveling on the road. However, if noise or the like is mixed, the difference image is disturbed, and the position and contour of the detected image are disturbed.
There is a problem that erroneous detection occurs. In addition, a vehicle newly entering the corner of the screen has a problem that the detection accuracy is deteriorated because only a part of the vehicle is photographed in the screen.

[0004]

The present invention employs the following structure to solve the above problems. <Structure 1> A camera that captures a moving object, a current image frame whose background is captured together with the moving object, a previous image frame whose background is captured together with the moving object immediately before capturing the current image frame, and a background that captures the background A storage unit for storing an image frame, a binarization unit for binarizing a difference between the background image frame and the previous image frame to obtain a binarized difference image, and a previous image frame and a binarized difference image, A template creating unit that obtains a template composed of an image of a small area including an image of each moving object, a motion vector detecting unit that detects a motion vector of each moving object by matching this template with the current image frame, Mask image creation for obtaining a mask image for masking a portion other than the image of each moving object in the current image frame from the outline of the template at the time of detection Wherein the template creating section masks unnecessary portions of the binarized difference image with a mask image received from the mask image creating section to obtain a template for each moving body. Detection device.

<Structure 2> In structure 1, a background block area is set in a part where the moving object enters the image, and the binarization unit generates a binarized difference image using the image of the previous image frame and the background block area. Then, the template creation section
A moving object behavior detecting device, wherein an unnecessary portion of a binarized difference image is masked with a mask image to create a template.

<Structure 3> In structure 1 or 2, the motion vector detecting section sets a predetermined detection area in the current image frame excluding a peripheral portion thereof, and performs matching using a template only in this detection area. A mobile object behavior detecting device characterized by performing the following.

<Structure 4> In the structure 1 or 2, the motion vector detecting section performs matching while expanding and contracting the template when moving the template in the moving direction of the moving object. .

<Structure 5> In the structure 1, when a difference is generated between the already stored background image frame and the previous image frame without the moving object, the storage unit stores the difference between the already stored background image frame and the previous image frame without the moving object. A moving object behavior detecting device for exchanging an image frame.

<Structure 6> In the structure 2, when a difference is generated between the image of the background block area already stored and the corresponding portion of the previous image frame without the moving object, the storage unit stores the image of the background block area already stored. A moving body behavior detecting device for exchanging an image with a corresponding portion of a previous image frame having no moving body.

<Structure 7> In the structure 5 or 6, the image of the background portion without the moving object in the previous image frame is extracted using the binarized difference image, and the previous image frame to be exchanged is extracted. A moving object behavior detecting device comprising a background image area creating unit for obtaining the moving object.

<Structure 8> In structure 1 or 2, when the template creating unit creates a template using the binarized difference image and the mask image already created and stored, the binarized difference A moving object behavior detecting device, wherein an internal dividing line between an area including a moving object of an image and an area including a moving object of a mask image is selected as a contour of a template.

[0012]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described below using specific examples. <Embodiment 1> FIG. 1 is a block diagram showing a specific example of a moving body behavior detecting apparatus according to the present invention. This device is camera 1, A
/ D conversion unit 2, current image frame storage unit 3, previous image frame storage unit 4, background image frame storage unit 5, difference units 6, 2,
It comprises a value conversion section 7, a template creation section 8, a motion vector detection section 9, and a vehicle area mask image creation section 10.

The camera 1 is fixedly arranged by a suitable frame (not shown) so that a vehicle can be photographed from a relatively high position on a road. The camera 1 photographs a running vehicle together with a road and outputs the photographed image as an analog signal. The A / D converter 2 is a device for converting an analog signal input from the camera 1 into a digital signal. The current image frame storage unit 3 is a memory for temporarily storing the digitized current image frame 13. The previous image frame storage unit 4 is a memory for temporarily storing a current image frame shot immediately before as a previous image frame 14. The background image frame storage unit 5 includes a memory that stores a background image frame 15 photographed in advance when there is no traveling vehicle on the road.

The differentiator 6 is a circuit that receives the previous image frame 14 from the previous image frame storage unit 4, receives the background image frame 15 from the background image frame storage unit 5, and outputs a difference value between the two. The binarizing unit 7 is a device that binarizes the output of the differentiator 6 based on the set threshold value. Thus, the binarizing section 7 outputs the binarized difference image 16. The template creating unit 8 is a device that creates a template to be described later and supplies the template to the motion vector detecting unit 9. The motion vector detection unit 9 is a device for detecting an image that matches the template of each vehicle in the current image frame 13 and obtaining a motion vector of the vehicle corresponding to the template. The vehicle region mask image creation unit 10 is a device that generates a vehicle region mask image suitable for detecting the next motion vector.

FIG. 2 shows a current image frame 13 stored in the current image frame storage unit 3 shown in FIG. 1, a previous image frame 14 stored in the previous image frame storage unit 4, and a background image frame storage unit 5. Background image frame 15 stored in
FIG. As shown in this figure, if the difference between the background image frame 15 and the previous image frame 14 is obtained, the images of the two vehicles in the previous image frame are extracted. By separating the image of each vehicle from the image and performing matching with the current image frame 13, it is possible to detect how much the vehicle has moved from the previous image frame.

FIG. 3 is an explanatory view of the output of the differentiator 6 shown in FIG. FIG. 4 shows the binarizing unit 7 shown in FIG.
FIG. The differentiator 6 shown in FIG. 1 receives the previous image frame 14 and the background image frame 15 and calculates the absolute value difference between the two. The differentiator 6 calculates a difference value between the previous image frame 14 and the background image frame 15 shown in FIG. 2 on a pixel basis or on a block basis described later, and binarizes the difference with a predetermined threshold. Thus, an absolute value difference block 19 as shown in FIG. 3 is obtained. Furthermore, a valuation block in which the isolated difference in which the periphery is considered to be the background does not become “0” is
It is preferable to obtain a clear image of the vehicle portion by performing noise elimination processing such as automatically replacing the block with the same block as the background.

As shown in FIG. 4, the output of the binarizing section 7 has a content in which only the actual vehicle image area 21 is left in the binarized difference image 16. In order to obtain such an image, the binarization unit divides the output of the differentiator into blocks of an appropriate size and integrates all the pixel values in the block as shown in FIG. 3, for example. Whether the entire block is set to "1" or "0" is performed depending on whether the integrated value is larger than a predetermined threshold value. This makes it possible to perform image processing in which the background image portion is all white and the vehicle existing portion is all black. The result is as shown in FIG.

FIG. 5 illustrates a case where the template creating section 8 shown in FIG. 1 generates a template based on the binarized difference image 16 output from the binarizing section 7. FIG.
4 accepts the binarized difference image 16 shown in FIG. 4 and specifies which part of the original image frame should be cut out for each vehicle to obtain an image of that vehicle. Individual vehicle area mask image 2
Accept 2-1 or 22-2. These are all shown in FIG.
As shown in (a) and (b), the vehicle area 23 is arranged in a window shape in a part of the background area 24. Such individual vehicle region mask images 22-1 and 22-2 are used for the traveling vehicle 1
Created for each car.

The region from which one vehicle is cut out in the binarized difference image is determined by the individual vehicle region mask image 22-1 shown in FIG. Thus, unnecessary signals are eliminated from the binarized difference image, and a template of one vehicle is cut out from the previous image frame. FIG. 5C shows an example of the template. After that, the motion vector detection unit 9 moves the template to perform matching with the current image frame 13 and detects a motion vector indicating the moving direction of the vehicle from the previous image frame to the current image frame. The vehicle area mask image creating unit 10 converts the position and outline information of the template after the motion vector detection into motion vector detecting unit 9.
From the individual vehicle area mask images 22-1 and 22
-2 is generated. In this way, it is possible to obtain a mask capable of completely eliminating unnecessary signals in an area other than the template.

FIG. 6 shows an operation flowchart of the template creating section. As shown in the figure, first, the template creating unit 8 receives the binarized difference image output from the binarizing unit 7 in step S1, and determines whether or not there is a vehicle here in step S2. If the data of the entire binarized difference image is all "0", there is no vehicle. In this case, the process ends. On the other hand, when it is determined that there is a vehicle, the process proceeds to step S3, and the individual vehicle mask image is accepted. Then, using these, a template is cut out from the previous image frame in step S4, and the template is output in step S5. Note that the vehicle area mask image creation unit 1
0 adds the number of templates used by the motion vector detection unit 9 and the number of vehicles that have newly entered after that, and generates an individual vehicle region mask image by adding the number.

FIG. 7 is a flowchart illustrating the operation of step S5 in FIG. 6 in more detail. First, the template creating section 8 labels the connected pixels of the binary difference image with each other (step S1). That is, a label is applied to each cluster of black pixels scattered in the binarized difference image 16 as shown in FIG. Then, in step S2, it is determined whether N is greater than "0". That is, it is determined whether a vehicle exists in the image.
If a vehicle exists, the process proceeds to step S3, and the parameter i is set to “0”. This parameter i is a counter that is added to each template when generating templates one by one.

In step S4, unnecessary signals are removed from the binarized difference image 16 using the individual vehicle area mask image.
That is, when the vehicle region in the individual vehicle region mask image and the label region existing in the binarized difference image 16 partially overlap, signals other than the label region are removed. In step S5,
A template is cut out from the previous image frame 14. That is, the binarized difference image 1 overlapped in step S4
6 is cut out from the previous image frame 14 only at the same position as that of the label area, and is used as a template. Next, when the processing is repeated by the number of vehicles in steps S6 and S7, templates can be generated for all vehicles. In step S8, of the labels remaining in the binarized difference image, only the labels at both ends of the image are determined to be vehicles that have newly entered, and a corresponding portion at the same position is cut out from the previous image frame 14, and a new image is obtained. Output as a template. This makes it possible to prevent a template from being created by determining a noise portion as a label. Through the above processing, templates are generated for all vehicles.

Next, an operation flowchart of the motion vector detecting section shown in FIG. 8 will be described. Step S1 in FIG.
This is a step for detecting a motion vector if there is at least one vehicle. In step S2, the parameters i and e are initialized. Then, in step S3, the vehicle template is accepted. Next step S
In step 4, a motion vector is detected between the current image frame and the template by pattern matching. That is, whether or not there is an image that matches the template in the current image frame is detected by a comparison process or the like. As shown in FIG. 2, for example, as shown in FIG. 2, if the vehicle gradually becomes smaller from the bottom of the screen to the top, matching is performed while expanding and contracting the template. Then, in step S5, the motion vector at the time of the best match is output.

When the position coordinates on the image of the template are detected by the matching, the position coordinates of the same template in the previous image frame 14 are compared with each other to determine the position of the vehicle from the previous frame to the current frame. It is possible to obtain a motion vector indicating how much distance has been moved in the direction. In the next step S6, it is determined whether or not the template at the time of the best match is out of the vehicle detection area. As will be described later, for example, if template matching is performed on the entire image, processing must be performed even on the peripheral portion of the image where only a part of the vehicle appears in the image.
In this case, erroneous detection may occur due to noise or the like. Therefore, not the entire image but only the central portion excluding the peripheral portion is set as a target of matching.

The vehicle detection area is a central portion thereof, and will be described later in more detail with reference to FIG. If the template deviates from this range, the vehicle is not targeted for behavior detection. Therefore, e prepared as a parameter for subtraction in step S8 is incremented. On the other hand, if the matching is performed in the vehicle detection area,
The template with the best match is output for post-processing or the like (step S7). Then, the parameter i is incremented in step S9, and the process proceeds to the next vehicle (step S10). Step S2 to step S10
Is repeated for the number of vehicles, and when the process is completed, in step S11, the total number of vehicles is reduced by the number of vehicles outside the vehicle detection area, and the process ends.

The above-described template matching process is performed only for a predetermined range in the image, or the next moving distance or direction is estimated from a plurality of past motion vectors for the same vehicle, and the template matching process is performed within that range. It is better to do it. This makes it possible to optimize the calculation amount and the calculation time of the matching process. In order to move the template to the range estimated from the past motion vector, for example, a plurality of past vectors for the i-th vehicle are set to mv
Assuming that (i, 1), mv (i, 2),..., Mv (i, n), the reference vector mv (i, 0) of Expression (1) is obtained. mv (i, 0) = Σ _{p = 1} ⁿ (α (p) × mv (i, p)) where α (p)> 0 and Σ _{p = 1} ⁿ α (p) = 1 (1) An appropriate range in front, back, left and right around mv (i, 0) may be set as the range for moving the template.

When the template is moved up and down, the template may be expanded or contracted by thinning out an appropriate amount of pixels according to the moved size to reduce or enlarge the template by interpolation. Good. In the matching process, the average of the absolute values of the pixel differences is obtained while moving the template between the current image frame and the template. When the average of the absolute values is minimum, it may be determined that the matching has been completed.

For example, when a predetermined vehicle detection area is set inside the screen, the value of the pixel inside the vehicle detection area is "1".
Then, a determination image in which the value of the pixel outside the vehicle detection area is set to “0” is generated. Then, the logical product of the corresponding pixels of the determination image and the individual vehicle region mask image is obtained. At this time, for example, “0” is set for a black pixel, and “1” is set for a white pixel. In this way, if the value of each pixel of the logical product image is all “0”, the vehicle in the individual vehicle region mask image is out of the vehicle detection region.
That is, when the vehicle region 23 shown in FIG. 5 in the individual vehicle region mask image exists inside the vehicle detection region, all pixels in this portion are “1”, and all other pixels are “0”.
Becomes On the other hand, when the vehicle region partially protrudes outside the vehicle detection region, the value of the pixel in this portion becomes “0”. Therefore, if all of them protrude, they are all "0", and it can be determined that the vehicle is out of the vehicle detection area. This means that if only a small part of the vehicle appears on the screen,
By making this a target of motion vector processing, it is possible to prevent erroneous detection from occurring.

<Effect of Specific Example 1> As described above,
According to the device of the specific example 1, a binarized difference image is obtained in advance by using the background image frame and the previous image frame, and a template from which unnecessary noise components are removed for each vehicle by using the individual vehicle region mask image. Since it is generated and used for motion vector detection, motion detection of each vehicle can be detected with high accuracy and without error. Also, in the matching process using the template, if the template is enlarged or reduced in accordance with the amount of movement of the template, a vehicle that is photographed larger as approaching in the image and smaller as moving away from the image with high accuracy. A motion vector can be detected. Also, by setting a vehicle detection area that is one size smaller than the image actually captured by the camera and excluding the peripheral part, and not detecting a motion vector for a vehicle outside the vehicle detection area, template matching can be performed. As a result, a target that causes a detection error is excluded, and highly reliable behavior detection can be performed. In addition, the apparatus of the present invention is not limited to the vehicle in the specific example 1 and other specific examples described below, and it is possible to take a picture with a camera, and it is necessary to detect a time-varying behavior. It can be implemented for any moving object.

<Embodiment 2> FIG. 9 shows a block diagram of the apparatus of Embodiment 2. This apparatus is different from the apparatus of the first embodiment in that a background block area image storage unit 29 for supplying a background block area image is connected to the differentiator 6. The other parts have the same configuration as that of FIG. 1 and duplicate description is omitted.

FIG. 10 is a view for explaining the processing of the background block area. FIG. 10A shows a screen 30 captured by a camera. The direction of travel of the vehicle is indicated by the left arrow. This road has two lanes, and each vehicle travels from the bottom of the screen toward the top of the screen. In this case, the area shown by the dashed line in the figure is the vehicle detection area 33.
Set to. As described above, the vehicle detection area 33 is a restricted area where a motion vector is to be detected when a template exists in this range. Accordingly, the vehicle detection area 33 excluding a predetermined width around the screen is excluded.
And On the other hand, in this example, a new vehicle is displayed on the screen 30.
Appear from the bottom of the screen. Therefore, in order to reduce the processing, in this specific example, the background block area 31
Is newly set, and for a new vehicle area, a template is created based on the difference in the background block area 31. Then, for the other vehicle regions, a template is created using the individual vehicle region mask image as described later.

(B) shows a background block area image 31-1 in which no vehicle exists. (C) shows a block 31-2 of the previous image frame in which a part of the vehicle exists. Further, (d) shows a block 31-3 of the current image frame photographed thereafter. If the background block area image 31-1 and the previous image frame block 31-2 are used, a binary difference image can be obtained in the same manner as in the specific example 1 described with reference to FIG. Images can be obtained.

In the specific example 2, the template creating section 8 performs an operation different from the extracting operation in step S5 in FIG. That is, the template creating unit 8 determines that the vehicle area of the individual vehicle area mask image (i) is the background block area 31
If it exists outside, the template is cut out from the previous image frame for this area as it is. The vehicle area of the individual vehicle area mask image (i) is the background block area 31
In the case where there are all of them, the template is cut out from the previous image frame for the same region as the label region of the binarized difference image as in the first embodiment.

On the other hand, when a part of the vehicle area of the individual vehicle area mask image (i) is included in the background block area 31, the template of the background block area 31 is changed to the binary difference image. It is cut out from the overlapped label area, the other part is cut out from the vehicle area of the individual vehicle area mask image (i), and both cut out templates are joined to be a template to be supplied to the motion vector detecting unit 9. The motion vector detection operation itself
Since the operation is exactly the same as that of the specific example 1, the duplicate description of the operation is omitted.

<Effect of Specific Example 2> As described above, for a vehicle that first appears on the screen, the background block area 31 is set to be limited to a part of the image, and a partial individual vehicle area mask image or the like is used using this. Has the effect of greatly reducing the amount of arithmetic processing. Further, if the processing is limited to a portion where the vehicle is largely captured as in the above example, further improvement in detection accuracy can be expected. In addition, in any case, there is an effect that the processing data amount is reduced and the calculation speed is improved.

The background block area can be set to various other areas. 11 and 12 show explanatory diagrams of examples of the background block area. For example, when the road is going up and down as shown in FIG. 11, the background block area 31 is set such that the lower part and the upper part of the screen are hatched. By setting the background block area 31 in a portion where the vehicle appears on the screen in this way, a newly entered vehicle can be extracted from both the upper and lower lines. FIG.
In the example shown in (1), the movement of the vehicle is detected using an image captured from the side of the vehicle. In this case, the background block area 31 is provided at the left end and the right end of the image. Thereby, a similar effect can be obtained.

<Embodiment 3> FIG. 13 is a block diagram of an apparatus according to Embodiment 3. This device is obtained by adding a background image frame creating unit 40 to the device shown in FIG. The other parts are exactly the same as those shown in FIG. The background image frame creation unit 40 is a unit that receives the previous image frame from the previous image frame storage unit 4, receives the binary difference image output from the binarization unit 7, and updates the background image frame.

FIG. 14 is a block diagram of the background image frame creating section. Here, two background image frame storage units 41 and 5 are provided. The background image frames used so far are stored in the background image frame storage unit 5. Meanwhile, the newly captured previous image frame is temporarily stored in the background image frame storage unit 41. Further, the output of the binarizing unit 7 shown in FIG. 13 is stored in the binarized difference image storage unit 43.

In this circuit, the background image frame and the previous image frame which have been used so far are processed by the differentiator 6 and the binarizing section 7 to obtain a binarized differential image. The binarized difference image is stored in the binarized difference image storage unit 43, and “0” is set when the content is, for example, all zeros, and “1” is set when the content is partially included in the selection switch unit 44. Supply. With this configuration, when the content of the previous image frame does not match the content of the background image frame, the content of the previous image frame newly stored in the background image frame storage unit is output from the selection switch unit 44. . As a result, the newly captured previous image frame is stored again in the background image frame storage unit 5 as a background image frame.

This processing may be performed when no vehicle is present in the previous image frame. However, when a vehicle is present, the background image may be rewritten only in a portion other than the image of the vehicle. When rewriting only a part, which part should be rewritten may be determined by using a binarized difference image.

<Effect of Specific Example 3> As described above, even when the background changes with the passage of time, all or part of the content of the background image frame storage unit is updated each time. The accuracy of detecting the motion vector of the vehicle can be increased. This effectively functions to accurately detect a vehicle even when a road image changes due to weather or other natural phenomena.

<Embodiment 4> FIG. 15 is a block diagram of an apparatus according to Embodiment 4. This device stores the background image frame creating unit 40 and the background image frame storing unit 5 of the device of the specific example 3 shown in FIG.
And a background block area image storage unit 29. The other parts are the same as in the device of the third embodiment. This background block area image creation unit 45
It has the same function as the background image frame creation unit 40 of the specific example 3 shown in FIG.

FIG. 16 is a block diagram of the background block area image creating section. As shown in the figure, the background block area image creating section 45 is composed of a background block area image storage section 46, a selection switch section 49, and a binarized difference image storage section 48. With this configuration, when the state of the background block area changes over time, the content can be updated, as in the specific example 3. That is, while the background block area image thus far is stored in the background block area image storage unit 29, a new image is taken from the previous image frame into the background block area image storage unit 46 and output from the binarization unit 7. The binarized difference image is taken into the binarized difference image storage unit 48, and the selection switch unit 49 is operated according to the contents. As a result, if the content of the background block area so far differs from the newly captured background block area, the content stored in the background block area image storage unit 29 is updated. Of course, some rewriting is also possible.

<Effect of Embodiment 4> As in Embodiment 3, when the background changes with the passage of time, the background block image is updated each time, and the vehicle motion vector can be detected with high accuracy.

<Embodiment 5> In the above-described embodiment, the template creation unit performs the following processing on the part where the divided vehicle region in the binarized difference image partially overlaps with the vehicle region in the individual vehicle region mask image. The template is created by cutting out the vehicle area in the binarized difference image. In this specific example 5,
A method of correcting a vehicle region in a binarized difference image with a vehicle region of an individual vehicle region mask image to create a template with higher accuracy will be described.

FIG. 17 is an explanatory diagram of a method for correcting a certain vehicle area in the binarized difference image. This process is performed, for example, according to the procedure shown in FIG. The vehicle area 51 of the individual vehicle area mask image output up to now from the vehicle area mask image creation unit 10 shown in FIG. 1 is shown by a solid circle in FIG. Further, the vehicle region 52 of the binarized difference image overlapping the vehicle region 51 is indicated by a hatched circle. In this case, when the two are superimposed, an internal segment 53 that internally divides the outline of both is obtained. The subdivision point may be the center of the corresponding point. This is indicated by a broken line in the figure. The circle corresponding to the internal dividing line 53 is used for cutting out a new template as the vehicle area 54 of the corrected individual vehicle area mask image.

In Example 1 of FIG. 17, the vehicle region 51 of the individual vehicle region mask image is wider than the vehicle region 52 of the binarized difference image. In Example 2, the relationship is reversed. In any case, the internal branch line 53 is obtained to obtain a vehicle area 54 of a new corrected individual vehicle area mask image of an appropriate size. In the case of Example 3, the positions of the vehicle region 51 of the individual vehicle region mask image and the vehicle region 52 of the binarized difference image are shifted. Even in such a case, the midpoints of the corresponding points on the contour line are plotted and connected to obtain the internal branch line 53. The internal dividing line 53 is not limited to the one that internally divides the vehicle regions 51 and 52 at equal intervals (1: 1). In addition, these vehicle regions 51 and 52 may be internally divided by arbitrary integers A and B.

<Effect of Specific Example 5> In this way, the template of the moving object can be more accurately cut out from the previous image frame. In each of the above embodiments, as described above, the target of the behavior detection is not limited to the vehicle, but can be applied to all persons, animals, and other moving objects. In addition, although the traveling direction of the vehicle is only one direction, the same method can be used to detect the behavior of a vehicle moving in various directions by photographing a road surface that is passing through the center line with a camera. . The signal for operating the selection switch unit 44 or the selection switch unit 49 of the specific example 3 or the specific example 4 is an average value of the difference output image or a weighted addition value αa + when α is a real number between 0 and 1. (1-α) b may be used. However, the pixel value is (a, b).

[Brief description of the drawings]

FIG. 1 is a block diagram of a moving object behavior detecting device according to the present invention.

2A is a current image frame, FIG. 2B is a previous image frame, and FIG. 2C is a background image frame.

FIG. 3 is an explanatory diagram of an output of a differentiator.

FIG. 4 is an explanatory diagram of an output of a binarizing unit.

FIG. 5 is an explanatory diagram of an individual vehicle area mask image.

FIG. 6 is an operation flowchart (part 1) of a template creation unit.

FIG. 7 is an operation flowchart (No. 2) of the template creation unit.

FIG. 8 is an operation flowchart of a motion vector detection unit.

FIG. 9 is a block diagram of an apparatus according to a specific example 2.

FIG. 10 is an explanatory diagram of processing of a background block area.

FIG. 11 is an explanatory diagram (part 1) of an example of a background block area.

FIG. 12 is a diagram (part 2) illustrating an example of a background block area.

FIG. 13 is a block diagram of an apparatus according to a third embodiment.

FIG. 14 is a block diagram of a background image frame creation unit.

FIG. 15 is a block diagram of an apparatus of a specific example 4.

FIG. 16 is a block diagram of a background block area image creation unit.

FIG. 17 is an explanatory diagram of correction of an individual vehicle area mask image.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Camera 3 Current image frame storage part 4 Previous image frame storage part 5 Background image frame storage part 6 Differentiator 7 Binarization part 8 Template creation part 9 Motion vector detection part

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Koji Matsumoto Inventor Oki Electric Industry Co., Ltd. 1-7-12 Toranomon, Minato-ku, Tokyo

Claims (8)

[Claims] 1. A camera for photographing a moving object, a current image frame photographing the background together with the moving object, a previous image frame photographing the background together with the moving object immediately before photographing the current image frame, and a background photographing. A storage unit that holds a background image frame, a binarization unit that binarizes a difference between the background image frame and the previous image frame to obtain a binarized difference image, A template creating unit that obtains a template composed of an image of a small area including an image of each moving object; a motion vector detecting unit that detects a motion vector of each moving object by matching the template with the current image frame; Mask image for obtaining a mask image for masking a portion other than the image of each moving object in the current image frame from the outline of the template when A moving unit, wherein the template forming unit masks unnecessary portions of the binarized difference image with the mask image received from the mask image forming unit to obtain a template for each moving object. Body behavior detection device. 2. The image processing apparatus according to claim 1, wherein a background block area is set in a part where the moving object enters the image, and the binarizing unit creates a binarized difference image from the image of the previous image frame and the background block area. The moving body behavior detection device, wherein the template creating unit creates a template for each moving body by masking an unnecessary portion of the binarized difference image with the mask image. 3. The motion vector detection unit according to claim 1, wherein the motion vector detection unit sets a predetermined detection area excluding a peripheral portion of the current image frame, and performs matching using a template only in the detection area. A moving object behavior detecting device characterized by performing the following. 4. The moving object behavior detecting device according to claim 1, wherein the motion vector detecting unit performs matching while expanding and contracting the template when moving the template in the moving direction of the moving object. 5. The storage unit according to claim 1, wherein when a difference is generated between the background image frame already stored and the previous image frame without the moving object, the storage unit stores the background image frame already stored and the previous image without the moving object. A moving body behavior detecting device characterized by exchanging a frame. 6. The image processing apparatus according to claim 2, wherein when a difference is generated between the image of the background block area already stored and the corresponding portion of the previous image frame without the moving object, the storage unit stores the image of the background block area already stored. A mobile object behavior detecting device for exchanging a part of a previous image frame with no mobile object. 7. The previous image frame to be exchanged according to claim 5 or 6, wherein an image of a background portion without a moving object in the previous image frame is extracted using the binarized difference image. A moving object behavior detection device comprising a background image area creation unit. 8. The binarized difference image according to claim 1, wherein the template creating unit creates a template using the binarized difference image and a mask image that has been created and stored. A moving object behavior detecting apparatus, wherein an internal dividing line between an area including the moving object of the mask image and an area including the moving object of the mask image is selected as a contour of the template.