JP2005050382A - Object detection method, device, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an intruding object detection method and an intruding object monitoring device for detecting a target object having intruded into an imaging range by reducing the erroneous detection of a moving object other than the target object. <P>SOLUTION: A differential image of a predetermined number of frames is obtained by respectively calculating differences on the basis of pixels from different image signals; a combined differential image is created by combining the predetermined number of differential images at a predetermined ratio; a binarized image is created by binarizing the combined differential image by a predetermined threshold value; and a binarized object in the binarized image is detected as an intruding object in a monitoring visual field. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to an object detection method, an object detection device, and an object detection program that detect an object using an imaging device, and in particular, detects an object within an imaging range in an environment where shaking such as trees and waves is observed. The present invention relates to an object detection method, an object detection device, and an object detection program for automatically detecting a target object to be detected from a video signal input from an imaging device.

  An intruding object monitoring device using an imaging device such as a camera as an image input means automatically detects an intruding object within the monitoring field of view or confirms the type of object, rather than being manually monitored by a conventional observer. Thus, a predetermined notification and warning treatment can be obtained. In order to realize such a system, first, an input image obtained from image input means such as a camera, a reference background image (that is, an image in which an object to be detected is not shown), and a time different from the input image are obtained. There is a method of comparing the obtained input image, obtaining a difference for each pixel, and extracting a region having a large difference as an object. This method is called a difference method and has been widely used. In particular, a method using a difference between an input image and a reference background image is called a background difference method, and a method using a difference between input images obtained at different times is called an interframe difference method.

  First, the background difference method will be described with reference to FIG. FIG. 5 is a diagram for explaining the processing principle of object detection in the background difference method, where 101 is an input image, 105 is a reference background image, 501 is a difference image by the background difference method, and 502 is a binarization of the difference image 501. An image, 112 is a subtractor, and 115 is a binarizer.

  In FIG. 5, a subtractor 112 calculates a difference in luminance value for each pixel of an image of two frames (in this figure, the input image 101 and the reference background image 105), outputs a difference image 501, and a binarizer 115. Indicates that the luminance value of each pixel of the difference image 501 is “0” when the luminance value is less than the predetermined threshold Th, and “255” is the luminance value of the pixel equal to or higher than the threshold Th (the luminance value of one pixel is 8 bits). As a calculation, a binarized image 502 is obtained.

  As a result, the humanoid object 503 shown in the input image 101 is calculated as a region 504 in which a difference is generated by the subtractor 112 and is detected as an image 505 having a luminance value “255” by the binarizer 115. . An example of applying the background subtraction method is, for example, Japanese Patent Laid-Open No. 9-288732.

  Next, processing of the interframe difference method will be described with reference to FIG. FIG. 6 is a diagram for explaining the processing principle of object detection in the inter-frame difference method, where 101 is a first input image, 102 is the same field of view as the first input image 101, and the first input image 101 is the same. A second input image obtained by imaging at different times, 601 is a difference image by the inter-frame difference method, 602 is a binary image of the difference image 601, 112 is a subtractor, and 115 is a binarizer. .

  In FIG. 6, the subtractor 112 calculates the difference between the luminance values for each pixel of the image of two frames (in this figure, the first input image 101 and the second input image 102) as in the case of FIG. The image 601 is output, and the binarizer 115 sets the luminance value of each pixel of the difference image 601 to a luminance value less than a predetermined threshold value “0” and equal to or more than the threshold value Th as in the case of FIG. The binarized image 602 is obtained by setting the luminance value of each pixel to “255” (the luminance value of one pixel is calculated by 8 bits).

  Thereby, the humanoid objects 603 and 604 shown in the first input image 101 and the second input image 102 are calculated as a region 605 in which a difference is generated by the subtractor 112, and the luminance value is calculated by the binarizer 115. It is detected as an image 606 of a cluster of “255”. An example of applying the interframe difference is, for example, Japanese Patent Registration No. 2633694.

  However, the background subtraction method has the feature that the target object can be detected even when the apparent moving speed of the target object on the input image is low, but when there is a moving object such as a leaf or wave. There is a problem that a moving object is erroneously detected. In addition, in the inter-frame difference method, when there is a moving object such as a tree leaf or a wave such as a wave, a time interval for acquiring a two-frame image to be subjected to difference processing is appropriately set (a leaf of a tree between two frame images). Although it is possible to reduce the detection of moving objects by mistake (false detection) by setting the change of the fluctuation of the wave and the wave to be small), the appearance of the target object to be detected on the input image There is a problem that the moving object is not detected as a target object when the moving speed is low.

  An object of the present invention is to provide an object detection method, an object detection device, and an object detection program for detecting a target object within an imaging range by reducing erroneous detection of a moving object other than the target object.

  In order to achieve the above object, the object detection method according to the present example is an object detection method for detecting an object in an image obtained from an imaging unit, and for a region far from the imaging unit, a background difference method The difference processing based on the interframe difference method is performed on the region where the distance from the imaging means is short.

  In order to achieve the above object, the object detection device according to the present example is an object detection device that detects an object in an image obtained from an imaging unit, and has a background for a region far from the imaging unit. Difference processing based on the difference method is performed, and difference processing based on the interframe difference method is performed on an area that is close to the imaging unit.

  In order to achieve the above object, the object detection program according to the present example is an object detection program for detecting an object in an image obtained from an image pickup unit, and is used for an area far from the image pickup unit. Difference processing based on the difference method is performed on a computer by performing difference processing based on the inter-frame difference method on an area having a short distance from the imaging unit.

The advantages and disadvantages of the interframe difference method and the background difference method are summarized as follows.
Advantages of inter-frame difference method: Set the time interval for acquiring two-frame images to be subjected to differential processing appropriately (set so that changes in leaves and wave fluctuations between the two-frame images are small). The detection rate can be reduced.

Disadvantages of inter-frame difference method: An object with a small apparent motion (a small amount of movement on the image at the time interval Δt) cannot be detected.
Advantages of the background subtraction method: It can detect objects with small apparent movements (it can also detect stopped objects).

Disadvantages of background subtraction method: Moving objects other than the target object to be detected are also detected.
As a result of an experiment (frame time interval Δt = 100 msec) in which the present inventors applied the interframe difference method and the background difference method to the detection of an ocean invasion monitoring ship, the following was found.

  If the inter-frame difference method is used, the reflection of the sunset reflected on the sea surface can be suppressed (the area of the false detection area is small even if it is detected). The background subtraction method cannot suppress erroneous detection of sunset reflection (the area of the erroneous detection area is large). There are many false detections of sunset reflections (because the waves appear larger the closer to the image). The inter-frame difference method cannot detect a distant ship (because the apparent movement amount is small).

  From these results, it can be concluded that the inter-frame difference method is effective when detecting the near side of the screen (near object), and the background difference method is effective when detecting the back of the screen (distant object). .

  Therefore, in the present invention, one feature is that the object detection performance is improved by applying a hybrid method in which the inter-frame difference method is applied in front of the screen of the television camera and the background difference method is applied in the back of the screen. It is a thing.

  According to the present invention, it is possible to provide an object detection method, an object detection apparatus, and an object detection program for detecting a target object within an imaging range by reducing erroneous detection of a moving object other than the target object.

Embodiments of the present invention will be described below with reference to the drawings. Like reference numerals refer to like parts throughout the drawings.
FIG. 4 is a block diagram showing a hardware configuration of an intruding object monitoring system to which the present invention is applied. First, the intruding object monitoring system will be described with reference to FIG.

  In FIG. 4, 401 is a television camera (hereinafter referred to as a TV camera), 402 is an image input I / F, 403 is a CPU, 404 is a program memory, 405 is an image memory, 406 is a work memory, and 407 is an output 1 / F. 408 is an image output 1 / F, 409 is a warning light, 410 is a monitor monitor, and 411 is a data bus.

  The TV camera 401 is connected to the image input I / F 402, the monitoring monitor 410 is connected to the image output I / F 408, and the warning lamp 409 is connected to the output I / F 407. The image input I / F 402, CPU 403, program memory 404, image memory 405, work memory 406, output I / F 407, and image output I / F 408 are connected to the data bus 411.

  In FIG. 4, the TV camera 401 captures an image within the imaging field including the monitoring target area. The TV camera 401 converts the captured image into a video signal, and inputs the video signal to the image input I / F 402. The image input I / F 402 converts the input video signal into image data in a format (for example, width 320 pix, height 240 pix, 8 bits / pix) that is handled by the intruding object monitoring system, and the image memory 405 via the data bus 411. Send to. The image memory 405 stores the image data sent from the image input I / F 402.

  The CPU 403 analyzes the image stored in the image memory 405 in the work memory 406 according to the operation program stored in the program memory 404. As a result of the above analysis, information indicating that an intruding object has entered the imaging field of view of the TV camera 401 is obtained. The CPU 403 displays, for example, a processing result image from the data bus 411 via the image output I / F 408 on the monitor monitor 410 and turns on the warning lamp 409 via the output I / F 407.

  The image output I / F 408 converts the signal from the CPU 403 into a format (for example, NTSC video signal) that can be used by the monitor monitor 410 and sends the converted signal to the monitor monitor 410. For example, the monitoring monitor 410 displays an intruding object detection result image.

FIG. 2 is a diagram showing a first processing flow of detection of an intruding object according to the present invention. This processing flow is executed using the hardware configuration of the intruding object monitoring system of FIG.
The first processing flow in FIG. 2 is a difference between the input image 101 input from the TV camera 401 in FIG. 4 and a previous input image of a predetermined number of frames (two frames or more) stored in the image memory 405. The image is calculated by the inter-frame difference method shown in FIG. 6, weighted and added to the obtained difference image of the predetermined number of frames, and the obtained synthesized difference image is binarized with a predetermined threshold value. This is a method for detecting an object that has entered the field of view of the TV camera 401.

  First, in an image input step 201, an input video signal imaged by the TV camera 401 is obtained as an input image 101 of 320 × 240 pixels (pix), for example. Next, in the frame counter clear step 202, the value of the frame counter, which is a variable for managing the image number that is the subject of the inter-frame difference, is set to 1.

  Next, in the inter-frame difference processing step 203, the input image 101 (in this case, expressed as a (x, y). (X, y) represents the pixel position on the image) and the image memory 405 are stored. The difference (denoted as ci (x, y)) from the previous input image (denoted as bi (x, y) here, i represents the value of the frame counter) is calculated.

  At this time, the input image stored in the target image memory 405 is determined by the frame number. For example, when the value of the frame counter is 1, it is stored most recently (for example, one frame before). The input image image (bi (x, y)). The difference for each pixel is calculated as follows.

Next, in the frame counter increment step 204, the value of the frame counter is incremented by one.
In the frame end determination step 205, if the value of the frame counter is less than the predetermined value N, the process branches to the inter-frame difference step 203. If the value is greater than or equal to the predetermined value N (for example, N = 3), the difference image composition step 206 Branch to Here, the predetermined value N represents the number of images used when performing the inter-frame difference with the input image 101, that is, the number of input images stored in the image memory 401. For example, when N = 4, The number of input images stored in the image memory 401 is four. In this case, four frames (ci (x, y), i = 1 to 4) are also obtained.

  Next, in the difference image synthesis step 206, the obtained N frame difference images are weighted by a predetermined weight coefficient image di (x, y) and added to obtain a synthesized difference image e (x, y). The composite difference image e (x, y) is calculated as follows.

However, the weighting coefficient image di (x, y) is

It sets beforehand so that it may become.
This weighted coefficient image di (x, y) represents how much each difference image ci (x, y) contributes to the combined difference image e (x, y). For example, dl (100, 100) = In the case of 255, it represents that the first difference image (cl (x, y)) has a contribution rate of 100% with respect to the synthesized difference image e (x, y) at the coordinates (100.100). (The weighted coefficient image is represented as an image of 8 bits per pixel. When the pixel value of the weighted coefficient image is 0, the contribution rate is 0%, and when the pixel value is 255, the contribution rate is 100%. Means).

  FIG. 14 is a diagram for explaining a difference image, a weighted coefficient image, and a combined difference image according to an embodiment of the present invention. FIG. 14 shows an example in which the number of frames of the difference image is 2, that is, ci (x, y) and i = 1,2. In order to simplify the description, the four pixel positions (1) to (4) of the difference image, the weighting coefficient image, and the combined difference image will be described. The luminance values of the differences of the pixel positions (1) to (4) of the difference image c1 (x, y) are output to the multiplier 113, and the pixel positions (1) to (4) of the difference image c2 (x, y) are output. ) Are output to the multiplier 113 '. In addition, the weighting coefficient image d1 (x, y) is given the weighting coefficient at the same pixel position (1) to (4) as the difference image by the value of the same dimension as the luminance value. For example, the weighting of the pixel position (1) The coefficient d1 (1) is 255, the weighting coefficient d1 (2) at the pixel position (2) is 127, the weighting coefficient d1 (3) at the pixel position (3) is 127, and the weighting coefficient d1 ( 4) is set to 0. Similarly, in the weighted coefficient image d2 (x, y), d2 (1) is 0, d2 (2) and d2 (3) are 128, and d2 (4) is 255. Therefore, multiplication is performed for each pixel by the multipliers 113 and 113 ′, and the multiplied output is added by the adder 114 and further divided by 255 to obtain a composite difference image e (x, y).

  Furthermore, the setting of the weighting coefficient image will be described with reference to FIGS. 7 and 8 are setting examples of weighting coefficient images when the present invention is applied to offshore monitoring. In FIG. 7, reference numeral 701 denotes an input image obtained by imaging the visual field range to be monitored, and FIG. 8 shows an example in which the predetermined value N is N = 4, and four weighting coefficient images di (x, y). ), I = 1 to 4 are displayed in an overlapping manner.

  In the example of FIG. 8, the sea surface area and other areas (breakwater and lighthouse areas) 804 are divided according to the value of the weighting coefficient, and the sea surface area is further divided into three areas according to the distance from the TV camera 401. It is divided into areas 801 to 803.

  Waves generated on the sea surface are observed to be larger as they are closer to the TV camera 401. For this reason, in the region close to the TV camera 401, the inter-frame difference must be performed so as to reduce the change in the luminance value due to the wave shaking. Therefore, it is necessary to shorten the image input time of two frames. That is, a difference image c1 (x, y) (two-frame image input time is short, for example, 100 msec) is used for the area 801 in front of the sea surface and the screen, and the area 802 separated from the TV camera 401 (for example, For an area 30 m or more away from the TV camera 401, an area further away from the TV camera 401 using the difference image c <b> 2 (x, y) (intermediate image input time interval of 2 frames, for example, 500 msec). For 803, for example, an area 100 m or more away from the TV camera 401, the difference image c3 (x, y) (the image input time interval of two frames is long, for example, 3 sec) is used.

However, in the region 804 where there is no wave shaking, the image input time interval of two frames can be increased, and therefore the difference image c4 (x, y) is used.
Accordingly, in the weighting coefficient image d1 (x, y), the pixels in the region 801 are set to 255, and the other pixels are set to 0.

  Similarly, in the weighted coefficient image d2 (x, y), the pixel in the area 802 is 255, the other pixels are 0, and in the weighted coefficient image d3 (x, y), the pixel in the area 803 is 255, and the other pixels are 0. In the weighting coefficient image d4 (x, y), the pixel in the region 804 is 255, and the other pixels are 0.

  This is illustrated as A to D in FIG. FIG. 11 is an example in which the pixel values of the weighted coefficient image di (x, y) are set to two values 0 and 255 in the scene of FIG. An image 1101 in FIG. 11A represents a weighted coefficient image d1 (x, y). In the weighted coefficient image d1 (x, y), the area 1101a and the area 1101b are set to a pixel value 255, and the other pixels are set to 0. . An image 1102 in FIG. 11B represents a weighted coefficient image d2 (x, y). In the weighted coefficient image d2 (x, y), the area 1102a and the area 1102b are set to a pixel value 255, and the other pixels are set to 0. . An image 1103 in FIG. 11C represents a weighted coefficient image d3 (x, y), and the weighted coefficient image d3 (x, y) sets the area 1103a to a pixel value 255 and sets the other pixels to 0. Furthermore, the image 1104 in FIG. 11D represents the weighted coefficient image d4 (x, y), and the weighted coefficient image d4 (x, y) sets the region 1104a to the pixel value 255 and sets the other pixels to 0.

  Of course, it is possible to set the weighting coefficient to a value smaller than 255 for pixels close to the boundary of each region. For example, for a pixel corresponding to the boundary between the region 801 and the region 802, d1 (x, y) = 127, It is good also as d2 (x, y) = 128. This is illustrated in FIG.

  FIG. 12 is an example in which the boundary width is 30 pix, and the pixel value of the image di (x, y) is set to three values 0, 127, and 255. Since the maximum value 255 of the pixel value is not divisible by 2, the remainder generated by the distribution of the weighting coefficient (contribution rate) is assigned to one of the weighting coefficient images. The pixel values 127 and 128 of the weighting coefficient image are assumed to be 127 here because they can be regarded as the same weighting coefficient without any difference of 0.4% with respect to the maximum weighting coefficient 255.

  An image 1201 represents a weighted coefficient image d1 (x, y). In the weighted coefficient image d1 (x, y), the area 1201a and the area 1201b are set to the pixel value 255, and the area 1201c and the area 1201d are set to the pixel value 127. Then, other pixels are set to 0. An image 1202 represents a weighted coefficient image d2 (x, y), and the weighted coefficient image d2 (x, y) sets a region 1202a (same as region 1201c) and a region 1202b (same as region 1201d) to a pixel value 128. , The region 1202c and the region 1202d are set to the pixel value 255, the region 1202e and the region 1202f are set to the pixel value 127, and the other pixels are set to 0. An image 1203 represents a weighted coefficient image d3 (x, y), and the weighted coefficient image d3 (x, y) sets a region 1203a (same as the region 1202e) and a region 1203b (same as the region 1202f) to a pixel value 128. , The region 1203c is set to a pixel value 255, and the other pixels are set to zero. Further, the image 1204 represents the weighted coefficient image d4 (x, y), and the weighted coefficient image d4 (x, y) sets the area 1204a to the pixel value 255 and sets the other pixels to 0.

  In this setting example, d4 (x, y) is represented by binary values of 0 and 255 as shown in FIG. 11D. This is not the characteristic that the area 804 becomes lower in the image (as the distance from the TV camera becomes smaller) like the other areas 801, 802, and 803, but the wave appears to be larger. This is because a difference between frames) or a background difference may be applied.

  In the example of FIGS. 11 and 12, the weighting coefficient of the weighting coefficient image is set as binary or ternary, but other weighting coefficient setting methods may be used. An example of the weighting coefficient setting method will be described with reference to FIG. FIG. 13 is an example in which the pixel values of each weighted coefficient image are assigned to 256 values from 0 to 255. An image 1301 represents the same scene as in FIG. 7, and a graph 1302 represents a contribution ratio distribution. In the graph 1302 representing the contribution ratio distribution, the position in the vertical direction corresponds to the y coordinate of the image 1301, and the width in the horizontal direction represents the contribution ratio (value of the weighting coefficient) to the composite difference image e (x, y). The graph 1302 is divided into three regions 1302a, 1302b, and 1302c, and the weighting coefficient images d1 (x, y) of the difference images c1 (x, y), c2 (x, y), and c3 (x, y), respectively. ), D2 (x, y), d3 (x, y). The region 1302a and the region 1302b are separated by a line connecting the point 1302g (corresponding to the y coordinate 220) and the point 1302h (corresponding to the y coordinate 80), and the region 1302b and the region 1302c correspond to the point 1302i (corresponding to the y coordinate 120). ) And a point 1302j (corresponding to the y coordinate 20). These points are set empirically according to the distance from the TV camera 401. For example, the point 1302g corresponds to a distance of 10m from the TV camera 401, the point 1302i corresponds to 30m, the point 1302h corresponds to a distance of 80m, and the point 1302j corresponds to a distance of 150m. It is set corresponding to the y coordinate on the image. When the areas are divided as shown in this figure, the widths of the areas 1302a, 1302b, and 1302c (weighting coefficients of d1 (x, y), d2 (x, y), and d3 (x, y)) are as follows.

Can be calculated as Here, for example, when the weighting coefficient at the position 1301a (y = 100) of the image 1301 is calculated, d1 (x, y) = 36 (width 1302d), d2 (x, y) = 168 (width 1302e), d3 (x , Y) = 51 (width 1302f). However, for a region 804 where there is no wave shaking (region where the background difference method can be applied), di (x, y) = 0 (I <4) and d4 (x, y) = 255. In this embodiment, areas 1302a, 1302b, and 1302c that determine the contribution ratio of the weighted coefficient image are divided using straight lines connecting the reference points 1302g, 1302h, 1302i, and 1302j, as in the graph 1302 of FIG. However, this may be divided using a curve.

  9 and 10 are setting examples of weighting coefficient images when the present invention is applied to outdoor monitoring. FIG. 9 shows an input image 901, and FIG. 10 shows an example in which N = 3, in which weighting coefficient images dl (x, y) and i = 1 to 3 are displayed in an overlapping manner. In this example, the area of the building, the ground and the sky is divided from the area of the vegetation, and the area of the vehicle tree is further divided into two according to the type of the vegetation.

  In the example of FIG. 9, the magnitude of the apparent movement on the image is such that the tree above the image is larger than the grass at the center of the image. In the region where the shaking is large, it is necessary to shorten the image input time interval of two frames for performing the inter-frame difference so as to reduce the change in the shaking of the trees. That is, the difference image c1 (x, y) is used for the tree region 1002, and the difference image c2 (x, y) is used for the grass region 1001. However, in the region 1003 where there is no shaking of the trees, the image input time interval of 2 frames can be increased, and therefore the difference image c3 (x, y) is used.

  Therefore, in the weighting coefficient image dl (x, y), the pixels in the region 1002 are set to 255, and the other pixels are set to 0. Similarly, in the weighting coefficient image d2 (x, y), the pixel in the region 1001 is 255, the other pixels are 0, d3 (x, y) is the pixel in the region 1003, and the other pixels are 0. Of course, as in the case of FIG. 7 and FIG. 8 described above, the weighting coefficient can be set to a value smaller than 255 in the region close to the boundary of each region. For example, for the pixel corresponding to the boundary with the region 1001 May be c1 (x, y) = 128 and c2 (x, y) = 127.

  Furthermore, as described in FIG. 13, the weighting coefficient image may be assigned with 256 values from 0 to 255. In FIG. 13, a weighting coefficient is assigned according to the distance from the camera 401. However, in the example of FIG. 9, the weighting coefficient is determined according to how much an object captured by the camera 401 is observed on the image. (In an area that is observed to move greatly (for example, 1002), it is observed with little movement so that the contribution ratio of d1 (x, y) with a short frame interval used in the difference method becomes high. In the region (for example, 1003), the frame interval used for the difference method is set to be long or the contribution rate of d3 (x, y), which is the difference from the back reference scene image, is set high).

  Since the setting of the weighting coefficient image only needs to be performed once at the time of installation of the intruding object monitoring system, the steps of setting the weighting coefficient image are not omitted in the flowcharts of FIGS.

  Next, in the binarization processing step 207 of FIG. 2, the composite difference image e (x, y) obtained in the difference image composition processing step 206 is used with a predetermined threshold Th (for example, Th = 20). Thus, the luminance value for each pixel of the composite difference image e (x, y) is 0 when the luminance value is less than the predetermined threshold Th, and the luminance value of the pixel that is equal to or higher than the threshold Th is 255 (the luminance value of one pixel is A binarized image f (x, y) is obtained as 8 bits).

  Next, in the intruding object determination step 208, whether or not there is a cluster of pixels having a luminance value of 255 in the obtained binarized image f (x, y) (the number of pixels is more than a predetermined number (for example, 100)). If there is a cluster of pixels having a luminance value of 255, it is regarded as an intruding object, branching to the alarm / monitor display step 210 in the branching step 209, and the pixel of the luminance value 255 If there is no block, the process branches to the input image storage step 211.

  In the alarm / monitor display step 210, the warning lamp 409 is turned on via the output I / F 407, or the monitoring processing result is displayed on the monitoring monitor 410 via the image output I / F 408, for example.

  Next, in the input image storing step 211, the input image 111 is stored in the image memory 405 as the input image b1 (x, y) one frame before. At that time, the input images b1 (x, y) to bN-1 (x, y) stored so far are copied from b2 (x, y) to bN (x, y), respectively. In this way, input images up to N frames before can be stored in the image memory 405.

By doing in this way, it can suppress that the moving object other than the target object which exists in the visual field of an imaging device appears in a difference image as a difference, and an exact intrusion object detection is attained.
FIG. 3 is a diagram showing a second processing flow of detection of an intruding object according to the present invention. FIG. 3 is obtained by adding a background difference step 301 and a reference background image update step 302 to the flow chart shown in FIG.

  In the background difference step 301, a difference for each pixel between the input image 101 and the reference background image 105 is calculated (this is assumed to be c (x, y)). In the difference image synthesis step 206, a difference image c (x, y) based on the background difference is synthesized instead of the Nth inter-frame difference image described in the processing flow shown in FIG. At this time, in the example of the first processing flow described above, the difference image c4 (x, y) four frames before is applied to the region 804 in FIG. The background difference image c (x, y) is applied.

  In the reference background image update step 302, for example, the pixels of the input image and the reference background image are averaged to obtain a new reference background image. The other steps are the same as the processing flow shown in the flowchart of FIG.

  Next, a flow of this series of processing will be described with reference to FIG. FIG. 1 shows an example in which three frames are used for the inter-frame difference and the background difference method is also used. In FIG. 1, an image 101 is an input image, an image 102 is an input image input at another time (for example, an input image one frame before), and an image 103 is an input image input at another time (for example, an input two frames before) The image 104 represents an input image (for example, an input image three frames before), and the image 105 represents a reference background image. The image 106 is a weighted coefficient image for the difference image between the input image 101 and the input image 102, the image 107 is a weighted coefficient image for the difference image between the input image 101 and the input image 103, and 108 is the input image 101 and the input image 104. , 109 represents a weighting coefficient image for the difference image between the input image 101 and the reference background image 105.

  In the input image 101 and the input image 102, the difference of each pixel is calculated by the difference unit 112-1, and the product of the difference image obtained as a result is calculated by the multiplier 113-1 for each pixel with the weighted coefficient image 106. , And input to the adder 114. In the input image 101 and the input image 103, the difference of each pixel is calculated by the differentiator 112-2, and the resulting image is obtained by calculating the product for each pixel with the weighted coefficient image 107 by the multiplier 113-2. , And input to the adder 114. In the input image 101 and the input image 104, the difference of each pixel is calculated by the difference unit 112-3, and the product of the difference image obtained as a result is calculated by the multiplier 113-3 for each pixel with the weighted coefficient image 108. , And input to the adder 114. The difference between the input image 101 and the background image 105 is calculated for each pixel by the difference unit 112-4, and the resulting difference image is a product for each pixel with the weighted coefficient image 109 calculated by the multiplier 113-4. , And input to the adder 114.

  The adder 114 adds the input four-frame difference image for each pixel to obtain a synthesized difference image 110. The obtained composite difference image 110 is compared with a predetermined threshold value for each pixel by the binarizer 115 and binarized with a luminance value equal to or higher than the threshold value 255 and a luminance value lower than the threshold value set to 0. An image 111 is obtained. By doing in this way, it can suppress that the moving object other than the target object which exists in the visual field of an imaging device appears in a difference image as a difference, and an exact intrusion object detection is attained.

  Therefore, according to the embodiment of the present invention, for moving objects other than the target object, such as leaves and waves, that exist within the monitoring range to be monitored, frames obtained from input images at different frame time intervals. By combining the difference image between the input image and the background difference image between the input image and the reference background image with a predetermined weight, it is possible to prevent moving objects other than the target object from appearing in the difference image as differences. The application range of the detection device can be greatly expanded.

The figure for demonstrating operation | movement of one Example of the detection process of the intruding object of this invention. The figure which shows the process sequence by 1st Example of this invention. The figure which shows the process sequence by the 2nd Example of this invention. The block diagram which shows the structure of one Example of the intrusion object monitoring system to which the intrusion object detection method of this invention is applied. The figure for demonstrating the processing principle of the object detection in the conventional inter-frame difference method. The figure for demonstrating the processing principle of the object detection in the conventional inter-frame difference method. The figure which shows one Example of the input image at the time of applying this invention to offshore monitoring. The figure which shows one Example of the weighting coefficient image at the time of applying this invention to offshore monitoring. The figure which shows one Example of the input image at the time of applying this invention to outdoor monitoring. The figure which shows one Example of the weighting coefficient image at the time of applying this invention to outdoor monitoring. The figure for demonstrating in detail the setting of the weighting coefficient image of FIG. The figure which shows one Example which sets the weighting coefficient of the weighting coefficient image of this invention by 3 values. The figure which shows one Example which sets the pixel value of the weighting coefficient image of this invention by multiple values. The figure for demonstrating the difference image of this invention, a weighting coefficient image, and a synthetic | combination difference image.

Explanation of symbols

101: input image, 102: 1 input image before frame, 103: input image before 2 frame, 104: input image before 3 frame, 105: reference background image, difference between frames of input image before 106: 1 frame Weighting coefficient image corresponding to the image, 107: weighting coefficient image corresponding to the inter-frame difference image based on the input image two frames before, 108: weighting coefficient image corresponding to the inter-frame difference image based on the input image three frames before, 109: Weighted coefficient image corresponding to the background difference image, 110: composite difference image, 111: binarized image, 112-1, 112-2, 112-3, 112-4: differentiator, 113-1, 113-2, 113-3, 113-4: multiplier, 114: adder, 115: binarizer.

Claims (3)

An object detection method for detecting an object in an image obtained from an imaging means,
For areas far from the imaging means, perform difference processing based on the background difference method,
For areas close to the imaging means, perform difference processing based on the interframe difference method,
An object detection method characterized by the above. An object detection device for detecting an object in an image obtained from an imaging means,
For areas far from the imaging means, perform difference processing based on the background difference method,
For areas close to the imaging means, perform difference processing based on the interframe difference method,
An object detection apparatus characterized by that. An object detection program for detecting an object in an image obtained from an imaging means,
For regions that are far from the imaging means, the difference processing based on the background difference method,
For regions close to the imaging means, difference processing based on the interframe difference method is performed.
An object detection program that is executed by a computer.