JP2003173435A - Moving body detecting method and moving body detecting device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a moving body detecting method and a moving body detecting device capable of detecting a moving body (vehicle and the like) with high accuracy regardless of the change of weather and the like. <P>SOLUTION: A detected area is picked up at first predetermined time intervals to acquire and store the picked-up images, and a binary image [i-n, i] obtained by binarizing the difference between the picked-up image [i-n] and the picked-up image [i], and a binary image [i, i+n] obtained by binarizing the difference between the picked-up image [i] and the picked-up image [i+n] are determined on the basis of the picked-up image [i] at a point of time [i], the picked-up image [i-n] in the past by n times with respect to the picked-up image [i], and the picked-up image [i+n] of in the future by n times with respect to the picked-up image [i]. A mask image [i] corresponding the point of time [i] is determined on the basis of the binary image [i-n, i] and the binary image [i, i+n], a moving body image [i] corresponding to the point of time [i] is determined on the basis of the mask image [i] and the picked-up image [i], and the moving body is detected on the basis of the information of the moving body image [i]. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a moving body detecting method and a moving body detecting device.

[0002]

2. Description of the Related Art A technique for detecting the behavior of a moving body (vehicle or the like) traveling on a road is required for the purpose of suppressing traffic accidents. For example, in Japanese Unexamined Patent Publication No. 10-105689, a preset detection area is photographed, and a background image (an image of a road when a moving body (vehicle, etc.) does not exist) and a current image (an image currently photographed) are taken. There has been proposed a moving body behavior detection device that detects a moving body (vehicle or the like) based on the difference between and.

[0003]

In the conventional method of detecting a moving body, the background is removed from the difference between the background image and the current image to extract only the moving body. However, in the conventional method of detecting a moving object, the influence of a change in the weather is large, for example, when the sun is shining and when the sun is hidden by clouds,
This may affect the detection accuracy of the moving body. In other words, when taking the difference between the background image in the daytime in fine weather and the current image in the daytime in cloudy weather, the amount of noise generated due to the difference in color or brightness is large, and moving objects should be detected accurately. Is difficult. To avoid this, it is preferable to capture the background image immediately before capturing the current image, but it is rare that the moving body (vehicle) is conveniently absent. The present invention has been devised in view of the above point, and provides a moving body detection method and a moving body detection device capable of accurately detecting a moving body (vehicle or the like) even with changes in weather or the like. With the goal.

[0004]

A first invention of the present invention for solving the above-mentioned problems is a method for detecting a moving body as set forth in claim 1. The moving object detection method according to claim 1, wherein the captured image [i-
n], the captured image [i], and the captured image [i + n],
Detects moving objects. Therefore, by using the three captured images that are very close in time to each other, the moving body (vehicle or the like) can be accurately detected with almost no influence of a change in weather or the like. Further, n is an integer of 2 or more, and for example, when an alarm or the like is generated according to the detected behavior of the moving body (vehicle or the like), a sampling interval (first predetermined time) necessary for generating the alarm or the like. The interval is maintained, and at the same time, the optimum sampling interval for detecting a moving object (in the case of n = 2, twice the first predetermined time interval, n = 3).
In this case, it is possible to set three times the first predetermined time interval or the like).

A second aspect of the present invention is a method for detecting a moving body as set forth in claim 2. In the moving body detection method according to claim 2, when the moving body is detected, the first
By shortening the predetermined time interval and increasing the first predetermined time interval when a moving object is not detected, unnecessary operation can be suppressed and the life of the power supply or the like can be extended.

The third invention of the present invention is a method for detecting a moving object as described in claim 3. In the moving object detection method according to claim 3, for example, when the distance to the moving object is long, or when the speed of the moving object is slow, n
The value of is increased to increase the sampling interval for detecting a moving object. Further, when the distance to the moving body is short, or when the moving body has a high speed, the value of n is reduced to shorten the sampling interval for detecting the moving body. Alternatively, the value of n is set to an appropriate value according to the position and speed of the moving body. In this way, it is possible to set the optimum sampling interval for detecting the moving body depending on the situation.

A fourth invention of the present invention is a method for detecting a moving object as described in claim 4. In the moving body detection method according to claim 4, by appropriately excluding a region in which an object that is not moving but is detected as a moving body (a flag swayed by the wind, trees, etc.) continuously exists. The moving body (in this case, the vibrating body) in the area can be determined as a non-moving body and excluded from the detection of the moving body. Therefore, the moving body (vehicle, etc.) can be accurately operated even when the weather changes.
Can be detected.

A fifth aspect of the present invention is a method for detecting a moving body as set forth in claim 5. In the moving body detection method according to claim 5, a region in which an object (a flag swayed by the wind, a tree, etc.) that is not moving but is detected as a moving body is almost constantly present is appropriately excluded. Then (even if it is intermittent, it can be excluded), it is possible to determine that the moving body (in this case, the vibrating body) in the area is a non-moving body and exclude it from the detection of the moving body. Therefore, the moving body (vehicle or the like) can be detected with higher accuracy even when the weather or the like changes.

A sixth aspect of the present invention is a method for detecting a moving body as set forth in claim 6. In the moving object detection method according to the sixth aspect, by appropriately setting the threshold value, the binarized image based on the difference between the two captured images can be easily and appropriately obtained.

Further, a seventh invention of the present invention is a moving object detecting method as described in claim 7. In the moving object detection method according to claim 7, when a binarized image is obtained,
For example, if there are many areas detected as moving objects, the threshold value is increased, and if there are few areas detected as moving objects, the threshold value is decreased. Objects that should not be can be properly excluded from moving object detection. Therefore, the moving body (vehicle or the like) can be detected with higher accuracy even if the weather or the like changes.

The eighth invention of the present invention is a method for detecting a moving body as described in claim 8. In the moving body detection method according to claim 8, in addition to the mask image [i] and the photographed image [i], a background image and a photographed image [i] when the moving body is not detected are used for detecting the moving body. ], The background-removed moving body image [i] is also used. For this reason,
The moving body (vehicle or the like) can be detected more accurately.

The ninth invention of the present invention is a moving object detecting apparatus as described in claim 9. By using the moving body detection device according to the ninth aspect, it is possible to realize a moving body detection device that can detect a moving body (vehicle or the like) with high accuracy even when the weather or the like changes.

The tenth invention of the present invention is the tenth aspect.
The moving body detection device as described in 1. Claim 1
If the moving body detection device described in 0 is used, the position and the moving direction of the detected moving body are obtained based on at least two moving body images, and the moving body (vehicle or the like) is located at a predetermined point (intersection or the like).
When a person is approaching, an alarm can be generated to call attention to people in the vicinity of the alarm device.

The eleventh invention of the present invention is the eleventh invention.
The moving body detection device as described in 1. Claim 1
If the moving body detecting device described in 1 is used, the moving body detecting device can be operated by using an appropriate power source. Further, when the solar cell is used, the energy cost for operating the moving body detection device can be reduced.

The twelfth invention of the present invention is the twelfth invention.
The vehicle approach warning device as described in 1. If the vehicle approach warning device according to claim 12 is used, it is determined whether or not the moving body is a vehicle based on the detected size of the moving body, and the determined vehicle is based on at least two moving body images. When the vehicle is approaching a predetermined point (intersection or the like) by determining the position and the moving direction of the vehicle, it is possible to suppress a traffic accident by issuing an alarm from an alarm device.

The thirteenth invention of the present invention is the thirteenth invention.
The vehicle approach warning device as described in 1. If the vehicle approach warning device according to the thirteenth aspect is used, the vehicle approach warning device can be operated by using an appropriate power source. Further, when the solar cell is used, the energy cost for operating the vehicle approach warning device can be reduced.

[0017]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 shows a schematic configuration diagram of an embodiment of a moving body detection apparatus to which the moving body detection method of the present invention is applied. In FIG. 1, the moving body C shows an alley β with poor visibility.
It shows an example of trying to cross the highway α from (is trying to go straight). In this example, there is an obstacle γ on the left side of the moving body C, and the main road α on the left side cannot be seen in the view from the moving body C. In the normal case, the driver of the mobile unit C slowly moves to the intersection and checks the traffic conditions on the left and right (of the main road α). Then, after traveling on the main road α and confirming that there is no moving body (vehicle or the like) coming toward the intersection, the vehicle crosses the main road α. However, in the case of dim light such as at dusk, the driver of the mobile body C may overlook the existence of the mobile body A traveling on the main road α and coming toward the intersection, which may lead to a traffic accident.

The image pickup apparatus 1 has a first predetermined time interval,
An image of the detection area (including the moving body A in FIG. 1) is captured. The processing device 2 stores the captured image in the storage device 3 at the first predetermined time interval, and processes the captured image stored in the storage device 3 according to the program to detect the moving body. Then, when the processing device 2 detects the moving body A traveling on the main road α and coming toward the intersection,
An alarm operation signal is sent to the alarm device 4. When the alarm device 4 receives the alarm operation signal from the processing device 2, the alarm device 4 generates an alarm by voice or light (various methods of alarm can be used). The driver of the moving body C can determine that another moving body (in this case, the moving body A) is approaching the intersection because the warning is issued from the warning device 4. As a result, the main road α can be safely crossed. The image capturing device 1, the processing device 2, the storage device 3, and the alarm device 4 may be separate devices, respectively.
It may be a device in which some of them are integrated (in the example of FIG. 1, the processing device 2 and the storage device 3 are integrated). The processing device 2 and the alarm device 4 may be connected by a wireless line or a wired line. Further, the installation place can be variously changed. As a power supply for driving each device, an AC power supply may be used, a solar cell may be used, or both an AC power supply and a solar cell may be provided and switched as needed. . Various power sources can be used as the power source. In the case where a solar cell is provided, for example, a battery panel is provided on the upper portion of the image capturing device 1 or the like.

[Detection Principle 1 of Mobile Object] Next, the detection principle of the mobile object detection method of the present invention will be described with reference to FIG.
The images used in this embodiment are all formed by orderly arranging individual pixels whose brightness (luminance) is converted into a digital amount (for example, vertical direction: 480 pixels, horizontal direction: 640 pixels). , Digital images. In addition, in each drawing, solid lines, dotted lines, one-dot chain lines, and two-dot chain lines are used as appropriate for the arrows connecting the images in order to clarify the flow of processing. First, the image capturing apparatus 1 captures an image of the detection area at a first predetermined time interval (for example, 1/8 second interval), and the processing apparatus 2
Stores the captured image in the storage device 3. In the example of FIG. 2, captured images [i-] are obtained at times [i-1], [i], and [i + 1].
1], the photographed image [i], and the photographed image [i + 1] are obtained and stored in the storage device 3. At time [i], the processing device 2 obtains the captured image [i] and stores it in the storage device 3, and further, based on the difference between the captured image [i-1] and the captured image [i]. Binarized image [i-1, i]
Is stored in the storage device 3. In this example, in the binarized image, the moving body is shown in black and the background excluding the moving body is shown in white (represented by binary values of black and white). Since the time [i-1] and the time [i] are very close to each other, there is almost no change in the brightness of the background or the like due to changes in the weather or the like, and the background can be appropriately removed. The details of how to obtain the binarized image will be described later.

At time [i + 1], the processing device 2
Obtains a captured image [i + 1], stores it in the storage device 3, and obtains a binarized image [i, i + 1] that is binarized based on the difference between the captured image [i] and the captured image [i + 1], It is stored in the storage device 3. Then, the binarized images [i-1, i] and 2
The mask image [i] is obtained from the logical product of the binarized image [i, i + 1]. Here, the mask image is represented by a binary value, and the mask image obtained at time [i + 1] is not the mask image corresponding to time [i + 1] but the time [i]
Is a mask image [i] corresponding to. The details of how to obtain the mask image will be described later. Then, the moving body image [i] is obtained based on the obtained mask image [i] and the captured image [i] corresponding to the time [i]. The moving object image [i] is obtained by extracting only the black portion of the mask image [i] from the captured image [i]. In the moving body image, the background other than the moving body is removed from the captured image, and only the moving body exists. Details of how to obtain the moving body image will be described later. In this example, the moving image [i] corresponding to the captured image [i] captured at the time [i] is extracted at the time [i + 1] instead of the time [i]. The interval of [i + 1] is short (1/8 second in this embodiment).
Therefore, it has almost no influence on the detection of a moving object and the generation of an alarm.

[Detection Principle 2 of Moving Object] FIG. 3 shows how the above-described processing of FIG. 2 is continuously executed at a first predetermined time interval. For example, time [i]
At the point of time, the captured image [i] is obtained and stored, and the binarized image [i-1, i] binarized based on the difference between the captured image [i-1] and the captured image [i] is stored. Seeking and remembering. Then, the binarized image [i-2, i-1] and the binarized image [i-
The mask image [i-1] is obtained from the logical product of
The moving object image [i-1] can be obtained by extracting the black portion of the mask image [i-1] from the captured image [i-1]. Similarly, at time [i + 1], the moving body image [i]
Can be obtained, and at time [i + 2], the moving body image [i
+1] can be obtained. Then, based on the moving body image [i-1], the moving body image [i], and the moving body image [i-1] at each first predetermined time interval, the moving direction and moving speed of the moving body, or the moving body You can calculate the distance to. The distance to the moving body can be estimated by the position of the moving body in each moving body image (in the present embodiment,
Mobiles move on the ground). The method for obtaining the moving direction, moving speed, position, etc. of the moving body will be described later.

[Method of Obtaining Binary Image] Next, a method of obtaining a binary image from the difference between two photographed images will be described with reference to FIG. In FIG. 4, the captured image is divided into 8 × 8 blocks (X, Y) for the sake of explanation. Also,
In this example, each block (X, Y) has a total of 1 × 4 × 4.
Six pixels are arranged in an orderly manner, and each pixel has a measured value based on brightness (luminance). For example, each pixel measures the brightness (luminance). As shown in FIG. 4, blocks (5, 2) and blocks (5, 3) of the captured image [i-1] and blocks (5, 5,) of the captured image [i].
The value of each pixel in 2) and the block (5, 3) is the amount of brightness (luminance) detected in each pixel (the portion indicated by the dotted line is shown with reference to the outer shape of the moving body). In addition, a portion in which the color is changed in each block is shown as a part of the moving body for reference.

Here, to obtain the difference between the photographed image [i-1] and the photographed image [i], the digital amount of each pixel of the photographed image [i] corresponds to the photographed image [i-1]. Subtract the digital amount of pixels. Then, the absolute value of the obtained difference is calculated. For example, the block (5,
The digital amount (6) at the upper right of the block (5, 3) of the captured image [i-1] is subtracted from the digital amount (12) of the pixel at the upper right of 3). Then, the absolute value of the difference is calculated for each pixel, and the calculated value is stored in the corresponding pixel portion. Next, the absolute value of the difference is binarized. For example, a pixel portion whose absolute value of difference is equal to or more than a threshold value (for example, “4” or more) is set to “1”, and a pixel portion whose absolute value of difference is less than the threshold value (for example, less than “4”). Is set to "0". This makes it possible to obtain a binarized image having only binary values of “1” and “0” from the difference between the two captured images. For example, set the pixel part of "0" to white,
When the pixel portion of "1" is set to black, the pixel shown in FIG.
The binarized image [i-1, i] is obtained. According to this method, it is possible to remove the background other than the moving object with high accuracy because the difference is between the two captured images at a very short time interval. However, since the pixels having a small difference in brightness (luminance) are regarded as the background and are removed, the brightness (luminance) of the moving body is substantially uniform, the moving speed of the moving body is low, or the shooting interval (first interval) is set. If the same moving body is superimposed on the captured image [i-1] and the captured image [i] (“moving body overlapping pixel” in FIG. 4), the moving body is present. However, it may be removed as a background. This phenomenon and the corresponding method will be described later.

[How to obtain mask image] Next, two 2
A method of obtaining the mask image from the logical product of the binarized images will be described. In the binarized image, each pixel is either "0" or "1" as shown in FIG. For each pixel, a logical product with the corresponding pixel is obtained. For example, the binarized image [i-1, i] and the binarized image [i, i +] shown in FIG.
1], the mask image [i] shown in FIG. 2 is obtained when, for example, the pixel portion of “0” is set to white and the pixel portion of “1” is set to black. To be

[Method of Obtaining Moving Object Image] Next, a method of obtaining a moving object image from a photographed image and a mask image will be described. As described above, in the mask image, each pixel portion is either "0" or "1". For example,
In FIG. 2, the pixel portion of the captured image [i] corresponding to the portion where the pixel of the mask image [i] is “0” (in this case, the white portion in the mask image [i]) is set to “0” ( In this case, it is white). In addition, the pixels of the captured image [i] corresponding to the portion where the pixels of the mask image [i] are “1” (in this case, the black portion in the mask image [i]) retain the digital amount as they are. In this way, the moving body image [i] can be obtained.

[Setting of First Predetermined Time Interval (1) (When Mobile Object is at Low Speed)] In FIG. 5, the moving speed of the mobile object is low, or the shooting interval (first predetermined time interval) is short. In this case, an example in which a moving object is superimposed on a binarized image obtained from the difference for each shooting time is shown. When the moving body is a vehicle, there are many areas where the brightness (luminance) is substantially uniform, and therefore, in the binarized image, a part of the overlapping portion of the moving body is erroneously recognized as the background and is removed (hereinafter, this state will be described). There is a possibility that it is referred to as a "middle-out state". For example, the captured image [i-
2] and the captured image [i-1], the binary image [i-2, i-1] obtained from the difference between the moving body A and the moving body B is overlapped with each other. The part has been removed and it has become white (there is a hollow state). Binary image [i
Similarly, the binarized image [i, i + 1] is in the void state.

The binarized images [i-2,
i-1] and the mask image [i-1] obtained from the binarized image [i-1, i] are only a part of the moving body as shown in the mask image [i-1] shown in FIG. Not represented. The moving body image [i-1] obtained from the mask image [i-1] and the captured image [i-1] is the moving body image [i-1] shown in FIG.
As shown in, only a part of the moving body is shown. The same applies to the moving body image [i]. Therefore, the first predetermined time interval is set to a time when the moving body to be detected does not overlap. In this case, the case where the moving body is extremely low speed is not assumed. This is because it is not necessary to detect it as a moving body and issue an alarm at extremely low speeds. For example, it is only necessary to be able to detect a moving body that moves at a speed of approximately 15 km / h or higher. First
The predetermined time interval of 1 is affected by the speed of the vehicle on the road in the detection area and the like, but is preferably 1/4 second or more. The setting of this value may be changed for each road in the detection area.

[Setting of First Predetermined Time Interval (2) (When Mobile Object is at High Speed)] FIG. 6 shows that the moving speed of the mobile object is high or that the interval between photographing (first predetermined time interval) is long. In this case, an example in which there are few captured images in which a moving body exists is shown. For example, as shown in FIG. 6, when the moving body exists only in the captured image [i], there is no moving body in the moving body image [i-1] and the moving body image [i + 1], and the moving body image exists. The moving body exists only in [i]. In this case, the moving direction and moving speed of the moving body cannot be obtained even by using the moving body images [i-1], [i], and [i + 1]. Therefore, the first predetermined time interval is set so that the number of captured images in which the moving body to be detected is large. The first predetermined time interval is influenced by the speed of the vehicle on the road in the detection area and the like, but is preferably ⅛ second or less. The setting of this value may be changed for each road in the detection area.

[Optimization of Time Interval for Detecting Low-speed or High-speed Moving Object] From the above description, the first method is to accurately detect the low-speed moving object using the difference between the captured images.
It is preferable to set the predetermined time interval of 1/4 seconds or more, and in order to accurately detect a high-speed moving object for each captured image, the first predetermined time interval is set to 1/8 seconds or less. Is preferred. A method for solving this will be described with reference to FIG. For example, as shown in FIG.
Take a picture every 8 seconds. However, the binarized image is obtained not from the captured image of the past once but from the captured image of the past n times (n is an integer of 2 or more). Similarly, the mask image and the moving body image are also obtained from images n times past. FIG. 7 shows an example in which “n = 2” is set (photographed images are photographed every 1/8 second). For example, at time [i + 1], the captured image [i + 1] is acquired and stored (twice in the past).
Captured image [i-1] and captured image [i]
Binarized image [i-
1, i + 1] is calculated and stored. Next, the obtained binarized image [i-1, i + 1] and the binarized image of the past twice [i-
3, i-1] and the mask image [i-1]. Then, the moving body image [i-1] can be obtained from the obtained mask image [i-1] and captured image [i-1]. In other words, in this case, the moving body image is obtained based on the difference between the photographed images every 1/4 second. In this case, the moving body image [i-1] corresponding to the captured image [i-1] captured at time [i-1] is extracted at time [i + 1] instead of time [i-1]. , The interval between the time [i-1] and the time [i + 1] is short (in this case, 1/4 second), so that it hardly affects the occurrence of an alarm.

At time [i + 2], the captured image [i + 2] is obtained and stored, and the difference between the captured image [i] (twice in the past) and the captured image [i + 2] (currently captured). The binarized image [i, i + 2] binarized based on is obtained and stored. Next, the obtained binarized image [i, i + 2]
And the mask image [i] is obtained from the logical product of the binary image [i-2, i] obtained twice. Then, the moving object image [i] can be obtained from the obtained mask image [i] and captured image [i]. Since the process of obtaining the moving body image is executed for each shooting, the time interval between the time [i + 1] and the time [i + 2] is 1/8 second. Therefore, the moving body image [i-
1], [i], and [i + 1] represent moving objects existing in the captured image every ⅛ second.

That is, a photographed image is obtained by photographing at the first predetermined time interval and stored, and the photographed image [i] at the time point [i] and the photographed image [n] past n times with respect to the photographed image [i]. i-n] (n is an integer greater than or equal to 2) and a captured image [i + n] n times in the future for the captured image [i], the captured image [i-n] and the captured image [i] are obtained. Binarized image [i−n, i] that is binarized based on the difference between the captured image [i] and captured image [i + n] i + n] is obtained, and the mask image [i] corresponding to the time point [i] is obtained and obtained based on the obtained binarized image [i−n, i] and the binarized image [i, i + n]. The moving object image [i] corresponding to the time point [i] based on the mask image [i] and the captured image [i],
The moving body is detected based on the obtained information of the moving body image [i]. In this example, the first predetermined time interval is ⅛ second and n = 2, and therefore the moving body is extracted every ⅛ second based on the difference every ¼ second. This makes it possible to accurately detect low-speed and high-speed moving bodies even with changes in weather and the like.

[How to obtain moving direction and moving speed of moving body] Next, a method of obtaining the moving direction and moving speed of the moving body based on the obtained moving body image will be described with reference to FIG. For example, as shown in FIG. 8, the obtained moving body image is divided into 8 × 8 blocks (X, Y). Also, in this example, each block (X, Y) is configured by a total of 16 4 × 4 pixels arranged in order, and each pixel contains a measurement value based on brightness (luminance). There is. By using the matching between the blocks, the destination block (X [i] of each block (X [i], Y [i]) in which the moving body exists is used.
+1], Y [i + 1]) is detected to obtain the moving direction and moving distance (motion vector) of the block. The moving speed can be obtained from the moving distance in a predetermined time.

Referring to FIG. 8, the motion vector plane [i-1, i] is calculated from the moving body image [i-1] and the moving body image [i].
A method for obtaining (motion vector of moving body from time [i-1] to time [i]) will be described. For example, the digital amount contained in each pixel in the block (5, 2) of the moving body image [i-1] has the value shown in FIG. Further, the digital amount contained in each pixel in the block (3, 4) of the moving body image [i] has the value shown in FIG. (The dotted line shows the outline of the moving body as a reference, and the portion in the block where the color is changed is shown as a part of the moving body.) In this case, the moving body image [i -1]
Block (5, 2) and the block (3, 4) of the moving image [i] are determined to match, and from the block (5, 2) to the block on the motion vector plane [i-1, i]. The vector going to (3, 4) is obtained. Similarly, "(block) motion vector" in each block
Then, the “(block) motion vector” in the similar direction and the similar distance is averaged to obtain the “(moving object) motion vector”. Then, the moving direction and the moving distance of the moving body can be obtained from the “(moving body) motion vector”. Further, the moving speed of the moving body can be obtained from the moving distance in a predetermined time. As a method for obtaining matching between blocks, there are (1) a sum of absolute values of difference gradation values of pixels in a block, (2) absolute value of a difference of average gradation values of pixels in a block, which are related to the related art. (3) Absolute value of difference between standard deviations of tone values of pixels in the block, (4) Cross-correlation coefficient of tone values of pixels in the block, (5) Matching by Fourier coefficient of tone values of pixels in the block There are various methods such as degree. In this embodiment, (1) the sum of the absolute values of the difference gradation values of the pixels in the block is used, but other methods may be used.

[Correction when a moving object is superimposed in the processing of a binarized image] Even if the method shown in FIG. 7 is used, when the moving object is moving at a slower speed, in the processing of the binarized image. However, there is a possibility that the moving body may overlap and a hollow state may occur. A correction method when a moving body is superposed will be described with reference to FIG. For example, when the moving body A and the moving body B are moving at a very low speed, as already described, the binarized image [i−2, i] and the binarized image [i, i + 2] are
A part of the overlapped portion is in a hollow state. The mask image [i] obtained from the binarized image in the void state cannot accurately represent the outer shape of the moving body, and the moving body image [i] obtained from the mask image [i] and the captured image [i]. Cannot detect a moving object accurately.

Therefore, the image when no moving object is present is set as the background image, and the background-removed moving object image [i] is obtained from the difference between the photographed image [i] and the background image. In this case, due to the difference in brightness (luminance) between the weather when the background image was captured and the weather when the captured image [i] was captured, the background-removed moving body image [i] is noisy. Ingredient is present. However, the background-removed moving body image [i] contains a noise component and a substantially accurate outer shape of the moving body. Then, the (corrected) moving body image [i] can be obtained by estimating the outer shape of the moving body from the overlapping portion of the background-removed moving body image [i] and the moving body image [i]. For example,
The overlapping portion of each of the isolated moving objects in the background-removed moving object image [i] and the moving object image [i] is estimated as a moving object and re-detected. Various methods can be used for the estimation method. In addition, as the background image, a captured image when the moving body image does not include a moving body when the moving body image is obtained is sequentially updated and used as a background image. Alternatively, the captured image when the moving body does not exist in the moving body image is stored together with the time, and the background image at the closest time is used. Various methods are conceivable as a background image storage method and a background image selection method.

[Shooting timing (first predetermined time interval), n value (“n (n is an integer greater than or equal to 1)” when obtaining the difference between the image n times past and n times future)) Change] In the above description, the first predetermined time interval is "1/8".
It was explained with the example that it is fixed to "second" and n is fixed to "2".
It can be changed according to the situation. [Change of First Predetermined Time Interval] For example, when the moving body image is obtained and there is no moving body in the moving body image, the first predetermined time interval may be lengthened (for example, 1
Seconds). When the distance to the moving body based on the position of the moving body in the moving body image is large (the moving body is far), the first
The predetermined time interval may be increased. Further, when the moving speed of the moving body obtained from the moving distance in the predetermined time is small, the first predetermined time interval may be lengthened. In the opposite case, the first predetermined time interval may be shortened. This makes it possible to set the optimum sampling interval for detecting the moving body. When the first predetermined time is changed, the calculation of the moving speed is also changed with the change of the first predetermined time interval.

[Change of n value] Further, for example, when the moving body image is obtained, if the distance to the moving body based on the position of the moving body in the moving body image is large (the moving body is far), n
The value may be increased (eg, n = 8). Alternatively,
If the moving speed of the moving body obtained from the moving distance in a predetermined time is small, the n value may be increased. In the opposite case, the n value may be reduced. Alternatively, the value of n may be set appropriately according to the position and speed of the moving body. This makes it possible to set the optimum sampling interval for detecting the moving body. When the n value is changed, the calculation of the moving speed is also changed with the change of the n value.

[Removal of Vibrating Body] Next, referring to FIG. 10, an object such as a flag swayed by the wind that is not moving but is detected as a moving body (in the present description, referred to as a vibrating body). ) Will be described. Note that in FIG. 10, the n value is set to “1” in order to easily understand the method of removing the vibrating body. In the present invention, a moving body such as a vehicle approaching a predetermined point (for example, the intersection shown in FIG. 1) is detected and an alarm is generated. Therefore, the vibrating body should not be detected as a moving body. However, since there is "motion" in each captured image captured at the first predetermined time interval, it is extracted as a moving body in the moving body image. For example, as shown in FIG. 10, when a vibrating body a (in this example, a flag swayed by the wind) is present in a captured image, there is “motion” in each captured image.
Therefore, the vibrating body a also exists in each binarized image, each mask image, and each moving body image. When the vibrating body a is present in the moving body image, the moving direction and moving distance of the block in which the vibrating body a is present are detected by the “(block) motion vector” shown in FIG. Therefore,
Before determining the “(block) motion vector”, a block for which the “(block) motion vector” should be obtained or a block for which the “(block) motion vector” should not be obtained is distinguished.

A method of distinguishing between a block for which a "(block) motion vector" should be obtained and a block for which it should not be obtained will be described with reference to FIG. FIG. 11 shows
Similar to FIG. 8, the obtained moving body image is divided into 8 × 8 blocks (X, Y). Then, an addition map is obtained from the moving body image divided into blocks. In the addition map, for example, "+2 (corresponding to a first predetermined value)" is assigned to a block in which a moving body exists, and "-1 (corresponding to a second predetermined value) in a block in which no moving body exists. Is assigned. It should be noted that the colored portions in the addition map are shown with reference to the area where the moving body exists. For example, in the addition map [i] obtained from the moving body image [i], blocks (2, 3), (3, 3) to (3, 6),
Since (4,3) to (4,6) and (5,3) to (5,6) have moving bodies, "+2" is assigned to the block and other blocks (blocks where moving bodies do not exist) are assigned. )
"-1" is assigned to. Then, the cumulative value (judgment value) of each block is stored in the cumulative map. It should be noted that the colored portions in the cumulative map are shown with reference to the area determined to be the vibrating body.

Here, whether the block for which the "(block) motion vector" should be obtained or the block for which it should not be obtained,
Two methods for distinguishing between will be described. [First Method] A first method is a method of obtaining a duration time in which a moving body exists for each block in each block obtained by dividing a moving body image. In the first method, the addition map and the cumulative map may be omitted. However, each block requires a duration map (not shown) for storing the duration in which the moving body existed. A block whose duration is equal to or longer than the second predetermined time is excluded from the determination of the moving body. In the example of FIG. 11, the moving body continuously exists in both the block (2, 3) and the block (3, 3). In both of these blocks, when the moving body is continuously present for a second predetermined time (for example, 5 seconds) or more, the block (2, 3) and the block (3, 3) are
It is determined that the “(block) motion vector” is a block that should not be obtained. A block other than the block in which the moving body exists is a block for which the “(block) motion vector” should be obtained. In the first method, when the moving body is no longer present, the determination exclusion of the moving body is restored (returning to the determination of the moving body).

[Second Method] In the second method, in each block obtained by dividing the moving body image, the first predetermined value is added when the moving body is present, and the second predetermined value is added when the moving body is not present. Is a method of subtracting the predetermined value of and obtaining the cumulative value (judgment value). When the cumulative value (judgment value) becomes equal to or larger than the third predetermined value, the block is excluded from the judgment of the moving body.
In the example of FIG. 11, the cumulative value (judgment value) of each block is stored in the cumulative map. For example, the cumulative map [i-
1] to the cumulative value (judgment value) of each block, the addition value (“+2” or “−” of each block of the addition map [i] is added.
1 ”) is added to obtain the cumulative map [i]. In this cumulative map, blocks whose cumulative value (determination value) is equal to or greater than a third predetermined value (for example, 50) are excluded from the determination of the moving body. In addition, the return from the determination exclusion of the moving body (returning to the determination of the moving body) is provided with hysteresis as shown in the “graph for each block” in FIG. It is shown that the body judgment is made) and the judgment is returned to 30 or less). Further, it is preferable to set appropriate upper and lower limit values (for example, upper limit value: 100, lower limit value: 0) as the cumulative value (determination value) in the cumulative map. (FIG. 11 is an example in which the lower limit: 0 is set.) The block excluded from the determination of the moving body is determined as the block for which the “(block) motion vector” should not be obtained. In addition, a block other than the block in which the moving body exists is a “(block) motion vector”.
Is the block to ask for. In addition, in both the first method and the second method, the first to third predetermined values can be changed according to the situation. For example, when there are many areas where the vibrating body exists (when the imaging device is shaking due to wind, etc.)
Can reduce the first predetermined value or increase the third predetermined value to reduce the determination area of the vibrating body.

[Removal of Other Noise Components] Next, with reference to FIG. 12, another method of removing noise components due to changes in the weather will be described. FIG. 12 is a diagram illustrating generation of noise due to snowfall and a method of removing the noise. For example, at the time of snowfall, when a captured image is captured as in FIG. 3, in addition to the moving object and the background, snow drifting in the air is captured in each captured image. Note that, in FIG. 12, in order to make it easier to understand how to remove other noise components,
The value is set to "1". Other noise components should not be detected as moving objects as well as the removal of the vibrating body described above. However, since there is "motion" in each captured image captured at the first predetermined time interval, it is extracted as a moving body in the moving body image. For example, as shown in FIG. 12, when another noise component (snowfall in this example) exists in the captured image, there is “motion” in each captured image. Therefore, other noise components are included in the binarized image, the mask image, and the moving body image (binarized image [i-2, i-1], [i-
1, i], mask image [i-1] and moving body image [i-
1]) also exists. When other noise components are present in the moving body image, the moving direction and moving distance of the block in which the other noise components are present are detected by the “(block) motion vector” shown in FIG. Therefore, other noise components are removed before obtaining the moving body image.

Now, two methods for removing other noise components before obtaining the moving body image will be described. [First Method] In the first method,
This is a method in which an isolated minute moving body is regarded as noise. As described above, the images used in this embodiment are all digital images in which individual pixels for measuring brightness (luminance) are arranged in an orderly manner. In addition, the moving image [i] has a pixel portion of the captured image [i] corresponding to a portion of the mask image [i] where the pixel is “0” (in this case, a white portion in the mask image [i]). The moving object image [i] can be obtained by setting it to "0" (in this case, white). From this, in the moving body image, it can be determined that the moving body is an isolated moving body by determining that the pixel is surrounded by the portion of “0”. Further, the determination as to be a minute moving body may be made by counting the number of pixels forming an isolated moving body and determining that the number is less than or equal to a predetermined number (for example, 6). By the above method, an isolated minute moving object (other noise component) can be removed from the moving object image.

[Second Method] In the second method, in the process of obtaining a binarized image, the threshold value for judging from the absolute value of the difference to the binary value "1" or "0" is appropriately changed. Two
This is a method of removing other noise components from the binarized image. The threshold value is the total area of regions (pixels) in which the absolute value of the difference in the binarized image is greater than or equal to the threshold value (the number of pixels of "1", etc.)
And the total area of the binarized image (the number of all the constituent pixels, etc.). In the case of weather such as snowfall and rainfall, the brightness (luminance) of the background and the like are closer to the measured values of snowfall and rain (noise component). The threshold may be changed according to the ratio of the total area of the regions (pixels) in which the absolute value of the difference in the binarized image is greater than or equal to the threshold to the total area of the regions (pixels) less than the threshold. . Various methods are possible for changing the threshold value based on the area of the region (pixel) in which the absolute value of the difference in the binarized image is equal to or greater than the threshold value.

For example, in FIG. 12, time [i-1]
In the binarized image [i-2, i-1] in, the threshold value is "4". In this example, the binarized image [i-2, i-
1] still has a noise component (snowfall in this case).
The ratio of the total area of the regions (pixels) in which the absolute value of the difference in the binarized image is equal to or more than the threshold value to the total area is about 12%. Here, from the “binarization threshold table” in FIG. 12, the threshold value (“4” in this case) corresponding to the area ratio of 12% is calculated as the next binarized image (binarized image [i -1, i]) is used in the process. Next, in the binarized image [i-1, i] at time [i], the threshold value is “4”. In this example, the binarized image [i-1, i]
A noise component (snowfall in this case) remains. Then, the ratio of the total area of the regions (pixels) in which the absolute value of the difference in the binarized image is equal to or more than the threshold value to the total area is about 25.
%. Here, "binarization threshold value table" in FIG.
Therefore, the threshold value (“6” in this case) of the portion corresponding to the area ratio of 25% is set to the next binarized image (binarized image [i,
i-1]) is used in the process.

Next, in the binarized image [i, i + 1] at time [i + 1], the threshold value is "6". Since the threshold value is increased from “4” to “6”, in this example, the noise component (snowfall in this case) is removed from the binarized image [i, i + 1]. Then, the ratio of the total area of the regions (pixels) in which the absolute value of the difference in the binarized image is equal to or more than the threshold value to the total area is about 25%. Here, from the “binarization threshold value table” in FIG. 12, the threshold value of the portion where the area ratio corresponds to 25% (“6” in this case)
To the next binarized image (binarized image [i + 1, i +
2]) is used in the process of obtaining. Next, in the binarized image [i + 1, i + 2] at time [i + 2], the threshold value is “6”.
Is. Since the threshold is “6”, in this example,
The noise component (snowfall in this case) is removed from the binarized image [i + 1, i + 2]. The ratio of the total area of the regions (pixels) in which the absolute value of the difference in the binarized image is equal to or more than the threshold value to the total area is about 15%. Here, from the “binarization threshold value table” in FIG. 12, the threshold value of the portion whose area ratio corresponds to 15% (“4” in this case)
To the next binarized image (binarized image [i + 2, i +
3]) is used in the process.

[Application to Vehicle Approach Warning Device] When the moving body obtained by the above moving body detection method is larger than a predetermined size (area), the moving body is recognized as a vehicle. Then, based on at least two moving body images,
Obtain the “(moving body) motion vector” of the recognized vehicle,
An alarm is issued from the alarm device based on the direction, speed, and position of the "(moving object) motion vector". For example, in the motion vector plane [i-1, i] in FIG. 8, the "(moving body) motion vector" is a predetermined point (for example, an intersection, and in this case, the motion vector plane [i-1, i]). If the direction is toward (close to) the bottom edge of], an alarm is generated. Alternatively, the size of the “(moving object) motion vector” that is (approaching) toward the predetermined point (in this case, the lower end of the motion vector plane [i-1, i]) is the predetermined size. If it is larger than that (when the speed is high), an alarm is generated. Alternatively, the position of the “(moving object) motion vector” heading (approaching) in the direction of the predetermined point (in this case, the lower end of the motion vector plane [i-1, i]) is set to the predetermined position. When it arrives (for example, when it reaches any one of the blocks (5, 1) to (8, 8)), an alarm is issued.

The moving body detecting method and the moving body detecting apparatus of the present invention are not limited to the processing or configuration described in the present embodiment, and various modifications can be made without departing from the scope of the present invention.
It can be added and deleted. For example, the configuration of the moving body detection device is not limited to that shown in FIG. 1, and various configurations are possible. The present invention is not limited to the detection of a moving body (vehicle) traveling on a road, but can be applied to the detection of various moving bodies. Although the present embodiment has been described using the luminance information,
It is also possible to use color information. (It is also possible to distinguish a similar group of color information from a moving body.) The numerical values used in the description of the present embodiment are examples, and the present invention is not limited to these numerical values. The images, tables, graphs, and the like shown in this embodiment are examples, and the present invention is not limited to these images, tables, graphs, and the like.
Further, the above (≧), the following (≦), the greater (>), the less than (<) and the like may or may not include an equal sign.

[0049]

As described above, the moving object detecting method according to any one of claims 1 to 8, the moving object detecting device according to any one of claims 9 to 11 and the vehicle approach warning device according to any one of claims 12 to 13 are provided. If used, a moving body (vehicle or the like) can be detected with high accuracy even when the weather changes.

[Brief description of drawings]

FIG. 1 is a schematic configuration diagram of an embodiment of a moving body detection device of the present invention.

FIG. 2 is a diagram for explaining the detection principle of the moving body detection method of the present invention.

FIG. 3 is a diagram showing how the process of FIG. 2 is continuously executed at a first predetermined time interval.

FIG. 4 is a diagram illustrating a method of obtaining a binarized image from a difference between two captured images.

FIG. 5 is a diagram showing an example in which a moving object is superimposed on a binarized image obtained from a difference for each shooting time.

FIG. 6 is a diagram showing an example in which there are few captured images in which a moving body exists.

FIG. 7 is a diagram illustrating a method of accurately detecting a low-speed moving body and a high-speed moving body with high accuracy.

FIG. 8 is a diagram illustrating a method of obtaining a moving direction and a moving speed of a moving body based on the obtained moving body image.

FIG. 9 is a diagram illustrating a correction method when a moving body is superposed.

FIG. 10 is a diagram illustrating an object (vibrating body) that is not moving but is detected as a moving body.

FIG. 11 is a diagram illustrating a method of removing an object (vibrating body) that is not moving but is detected as a moving body.

FIG. 12 is a diagram illustrating another method of removing a noise component due to a change in weather or the like.

[Explanation of symbols]

1 Imaging device 2 processing equipment 3 storage devices 4 Alarm device

─────────────────────────────────────────────────── ─── Continuation of front page (51) Int.Cl. ⁷ Identification code FI theme code (reference) G06T 7/60 150 G06T 7/60 150B G08B 21/00 G08B 21/00 U G08G 1/04 G08G 1/04 DF Term (Reference) 5B057 AA16 BA29 BA30 CA08 CA12 CA16 CB06 CB12 CB16 CH01 CH11 DA07 DB02 DB09 DC22 DC32 5C086 AA54 BA22 CA28 CB36 DA08 5H180 AA01 BB03 BB04 FA06 FA04 FA04 LL06L04 LL04L04 LL04L02 LL04L02 LL04L02 LL04L02 LL04L04L02 GA08 GA10 GA19 GA51 HA03 LA05

Claims (13)

[Claims] 1. A detection region is photographed at a first predetermined time interval, a photographed image is obtained and stored, and the photographed image [i] at time [i] and n times past the photographed image [i]. The captured image [i-n] (n is an integer of 2 or more) and the captured image [i + n] n times in the future with respect to the captured image [i] are used, and the captured image [i-n] and the captured image [ 2] based on the difference between the binarized image [i−n, i] and the photographed image [i] and the photographed image [i + n].
The binarized image [i, i + n] that has been binarized is obtained, and the obtained 2
The mask image [i] corresponding to the time point [i] is obtained based on the binarized image [i−n, i] and the binarized image [i, i + n], and the obtained mask image [i] and captured image A moving body detection method for obtaining a moving body image [i] corresponding to a time point [i] based on [i] and detecting the moving body based on information of the obtained moving body image [i]. 2. The moving body detection method according to claim 1, wherein when the moving body is detected, the first predetermined time interval is shortened,
A method of detecting a moving body, wherein the first predetermined time interval is lengthened when the moving body is not detected. 3. The detection area is photographed at a first predetermined time interval, the photographed image is obtained and stored, and the photographed image [i] at time [i] and n times past the photographed image [i]. The captured image [i-n] (n is an integer greater than or equal to 1) and the captured image [i + n] n times in the future with respect to the captured image [i] are used. 2] based on the difference between the binarized image [i−n, i] and the photographed image [i] and the photographed image [i + n].
The binarized image [i, i + n] that has been binarized is obtained, and the obtained 2
The mask image [i] corresponding to the time point [i] is obtained based on the binarized image [i−n, i] and the binarized image [i, i + n], and the obtained mask image [i] and captured image The moving body image [i] corresponding to the time point [i] is obtained based on [i], the moving body is detected based on the obtained information of the moving body image [i], and the detected position of the moving body is detected. And a velocity, the value of n is changed based on at least one of the two. 4. The moving body detection method according to claim 1, wherein the moving body image is divided into a plurality of areas, the moving body is determined based on information of each area, and the same moving body image is detected. A moving body detection method, wherein when a moving body in a region is continuously detected for a second predetermined time or longer, the moving body in the region is determined as a non-moving body. 5. The moving body detection method according to claim 1, wherein the moving body image is divided into a plurality of areas, the moving body is determined based on information of each area, and each area is determined. Is provided with a judgment value, the first predetermined value is added to the judgment value of the area where the moving body exists, and the second predetermined value is subtracted from the judgment value of the area where the moving body does not exist, The moving body detection method, wherein when the determination value is equal to or larger than the third predetermined value, the moving body in the area is determined as a non-moving body. 6. The moving object detection method according to claim 1, wherein the captured image is composed of a plurality of regions, and when a binarized image is obtained, correspondence between the two captured images is provided. A moving object detection method in which a binarized image is obtained by classifying into two types of areas, an area in which the absolute value of the difference in the brightness data of the area is greater than or equal to a threshold and an area in which the absolute value is less than the threshold. 7. The moving object detection method according to claim 6, wherein when a binarized image is obtained, the absolute value of the difference between the brightness data of the corresponding areas of the two captured images is equal to or more than a threshold value. A moving object detection method, wherein a threshold value is changed according to the area of a region. 8. The moving body detection method according to claim 1, further comprising setting a captured image when a moving body is not detected as a background image, and setting the captured image [i] as The background-removed moving object image [i] at the time point [i] is obtained from the difference with the background image, and based on the mask image [i], the captured image [i], and the background-removed moving object image [i], A moving object detection method of obtaining a moving object image [i] corresponding to [i]. 9. An imaging device, a processing device, and a storage device, wherein the imaging device captures an image of the detection region at a first predetermined time interval, and the processing device captures an image at a first predetermined time interval. A moving body detection device that stores the captured image in a storage device and detects a moving body using the moving body detection method according to claim 1. 10. The moving body detection device according to claim 9, further comprising an alarm device, wherein the processing device is based on at least two moving body images,
A moving body detection device which obtains the detected position and moving direction of the moving body and issues an alarm from an alarm device when the moving body is approaching a predetermined point. 11. The moving body detection device according to claim 9 or 10, comprising at least one of an AC power supply and a solar cell as a power supply. 12. An imaging device, a processing device, a storage device, and an alarm device, wherein the imaging device captures an image of the detection area at a first predetermined time interval, and the processing device has a first predetermined time interval. Then, the captured image is stored in a storage device, the moving body is detected using the moving body detection method according to claim 1, and the moving body is detected based on the size of the detected moving body. Is a vehicle, and based on at least two moving body images,
A vehicle approach warning device that obtains the determined position and moving direction of a vehicle and issues an alarm from an alarm device when the vehicle is approaching a predetermined point. 13. The vehicle approach warning device according to claim 12, comprising at least one of an AC power supply and a solar battery as a power supply.