JP2013148989A - Scan window size determination device, detection frame size histogram information acquisition device, and scan window size determination method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To determine a scan window size when an object such as a person is detected from an image photographed with a monitoring camera according to the size of the person or the like photographed with the monitoring camera.SOLUTION: Based on a detection frame size histogram information storage part 103 which holds statistical information on detection frame size for every plurality of areas of an image divided into the plurality of areas, and the statistical information stored in the detection frame size histogram information storage part 103, a scan window size of the areas is determined to prevent erroneous detection.

Description

  The present invention divides an input image into several regions in still image object detection, and determines a scan window size for determining a scan window size for still image object detection using detection frame size histogram information acquired for each region in advance. The present invention relates to an apparatus, a scan window size determination method, and a false detection removal apparatus and false detection removal method for removing false detection using the detection frame size histogram information acquired in advance.

  In still image object detection, in order to detect objects of various sizes in the image, it is necessary to resize the input image multiple times and perform detection processing. However, there are problems such as increased processing time and increased false detection. There is. As a conventional method for dealing with these problems, when an imaging device with an autofocus function is used, the object size in the input image is estimated and input according to the focus distance information and the angle of view information on the object. There are a scan window size determination apparatus and a scan window size determination method that perform still image object detection processing on the entire input image without performing resize processing of an image multiple times and using an estimated object size as the only scan window size. As a result, the conventional method increases the processing speed and suppresses erroneous detection of a size different from the assumed object size.

JP 2006-025238 A

  However, since the conventional scan window size determination apparatus and scan window size determination method use only the in-focus distance with the object, a pan-focus monitoring camera having a deep depth of field (the in-focus distance is wide), etc. When used, there is a problem that an optimal scan window size cannot be obtained.

  Further, since the object detection process is performed with the estimated single scan window size without performing the resizing process a plurality of times, there is a problem that the detection performance of an object other than the focused object is deteriorated.

  The scan window size determining apparatus according to the present invention includes a detection frame size histogram information storage unit that holds statistical information of a detection frame size for each region of an image divided into a plurality of regions, and a detection frame size histogram information storage unit. A scan window size determining unit that determines a scan window size of the area based on the stored statistical information.

  According to the above configuration, since an optimum scan window size can be determined for each divided image area, there is no need to detect an object with an unnecessary scan window size, and still image object detection processing is performed. It can be desired to speed up.

  The scan window size determining apparatus of the present invention includes an image input unit that inputs an image and an object detection unit, and the image input unit divides the input image into the plurality of regions, and detects the object. The unit is characterized in that an object is detected for each of the plurality of regions using the scan window size determined for each of the plurality of regions by the scan window size determining unit.

  According to the above configuration, similarly, since the scan window size can be determined for each divided image area, there is no need to detect an object with an unnecessary scan window size, and still image object detection is performed. It is possible to increase the processing speed.

  Furthermore, the scan window size determining apparatus of the present invention is characterized in that the detection frame size histogram information is an average value and variance of detection frame sizes when an object is detected for each region.

  According to the above configuration, the scan window size for each of the plurality of regions can be determined based on the detection frame size histogram information.

  Furthermore, the scan window size determining apparatus of the present invention is characterized in that the scan window size determining unit determines the scan window size using weighted sampling from the statistical information.

  According to the above configuration, the scan window size can be determined from the statistical information using weighted sampling.

  Furthermore, the detection frame size histogram information acquisition device for generating statistical information of the detection frame size stored in the detection frame size histogram information storage unit according to the present application includes: an image input unit that divides an input image into a plurality of regions; An object detection unit that detects an object in the input image using a scan window size; and the detection frame size of the region where the center of gravity of the obtained object detection frame is located when the object detection unit detects the object A detection frame size histogram voting unit for voting the detection frame size of the detection frame with respect to the histogram.

  With this configuration, statistical information used in the scan window size determining apparatus can be acquired for each divided region in the image.

  Furthermore, the scan window size determining apparatus according to the present application uses a detection frame size histogram information storage unit that holds statistical information of a detection frame size for each of the regions divided into a plurality of regions, and a plurality of scan window sizes. An object detection unit that detects an object in the input image, and when the object detection unit detects an object, the detection frame size histogram of the region where the center of gravity of the obtained object detection frame is located, And an erroneous detection removal unit that determines whether or not the object detection frame is erroneously detected.

  With this configuration, when an object is detected, an object detection frame having an object detection frame size that is difficult to detect statistically in each region in the input image can be removed as a false detection.

  According to the present invention, when detecting an object from an input image, the detection frame size can be determined for each region into which the input image is divided. Therefore, when detecting by changing the size of the detection frame in various ways, When there is no waste in detecting with a detection frame that does not exist in the area, and an object that does not exist in the divided area is detected, it can be determined as a false detection.

1 is a block diagram showing a schematic configuration of a scan window size determining apparatus according to Embodiment 1 of the present invention. FIG. 1 is a block diagram showing a detection frame size histogram information acquisition apparatus according to Embodiment 1 of the present invention. The flowchart which shows control of the scan window size determination apparatus which concerns on Embodiment 1 of this invention. The flowchart which shows control of the detection frame size histogram information acquisition apparatus which concerns on Embodiment 1 of this invention. Explanatory drawing of the histogram which concerns on Embodiment 1 of this invention Explanatory drawing of a stationary object detection using the scan window size determination measure which concerns on Embodiment 1 of this invention Explanatory drawing of the scan window size selection step by weighted sampling according to Embodiment 1 of the present invention. Explanatory drawing of the weighted sampling according to calculation of the detection frame size histogram which concerns on Embodiment 1 of this invention Explanatory drawing of the detection frame size histogram information acquisition apparatus which concerns on Embodiment 1 of this invention. The block diagram which shows schematic structure of the false detection removal apparatus which concerns on Embodiment 2 of this invention. The flowchart which shows control of the false detection removal apparatus which concerns on Embodiment 2 of this invention. Operation explanatory diagram of the false detection removal device according to Embodiment 2 of the present invention

  DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, preferred embodiments for carrying out the invention will be described in detail with reference to the drawings.

(Embodiment 1)
Embodiment 1 of the present invention will be described below with reference to the drawings.

  FIG. 1 is a block diagram showing a schematic configuration of a scan window size determining apparatus according to Embodiment 1 of the present invention. The scan window size determining apparatus 101 according to the present embodiment is configured using a normal computer. That is, it is configured by a computer including a volatile memory such as a CPU and a RAM, a nonvolatile memory such as a ROM, and a large-capacity storage device such as a hard disk or an SSD (Solid State Drive). Needless to say, a dedicated hardware configuration other than the configuration using the computer may be used.

  In FIG. 1, a scan window size determining apparatus 101 according to Embodiment 1 of the present invention includes an image input unit 102, a detection frame size histogram information storage unit 103, a scan window size determining unit 104, and an area object detecting unit 105. With.

  The image input unit 102 divides the input image into an arbitrary number of regions. The detection frame size histogram information storage unit 103 stores a histogram of detection frame size for each region of the image divided into a plurality of numbers. The scan window size determination unit 104 extracts an arbitrary number of scan window sizes for each region by weighted sampling with respect to the detection frame size histogram of each region read from the detection frame size histogram information storage unit 103. The area object detection unit 105 performs still image object detection processing on each area using an arbitrary number of scan window sizes extracted for each area by the scan window size determination unit 104.

  FIG. 2 shows a detection frame size histogram information acquisition apparatus 201 that acquires histogram information stored in the detection frame size histogram information storage unit 103.

  The detection frame size histogram information acquisition apparatus 201 includes an image input unit 102, a screen object detection unit 202, a detection frame size histogram voting unit 203, and a detection frame size histogram information storage unit 103. The frame size histogram information storage unit 103 is the same as that of the scan window determination apparatus of FIG.

  The image input unit 102 divides the input image into an arbitrary number of regions. The screen object detection unit 202 performs still image object detection processing on the entire input image with a predetermined start scale, scale magnification, and number of resizings on the input image input by the image input unit 102, The desired object is detected. The detection frame size histogram voting unit 203 obtains a region to which the center of gravity of the object detection frame obtained by the screen object detection unit 202 belongs, and the object detection frame obtained by the screen object detection unit 202 in the detection frame size histogram unique to that region. Vote for size. The detection frame size histogram information storage unit 103, after the processing of the screen object detection unit 202 and the detection frame size histogram voting unit 203 is completed for the entire input image, the detection frame size histogram of each region, its average value, standard Memorize the deviation.

  FIG. 3 is a flowchart showing the control of the scan window size determining apparatus 101 according to the first embodiment of the present invention.

  In parameter setting step S301, the number and size of areas into which the input image is divided, the detection frame size sampling number n from the detection frame size histogram of each area, and the like are defined, and the process proceeds to step S302.

  An image input step S302 inputs an image. At this time, the input image is divided into regions defined in step S301, and the process proceeds to step S303.

  In the detection area changing step S303, an area to be subjected to the still image object detection process is changed in each area in the image defined in step S301, and the process proceeds to step S304.

  In the detection frame size histogram information reading step S304, the detection frame size histogram information of the detection area defined in step S303 is read, and the process proceeds to step S305.

  In the scan window size selection step S305 by weighted sampling, the detection frame defined in step S301 is performed by weighted sampling based on the appearance probability of each detection frame size with respect to the detection frame size histogram of the attention area defined in step S303. A scan window size of size sampling number n is selected, and the process proceeds to step S306.

  In the scan window size changing step S306, one scan window size for scanning the detection region defined in step S303 is selected from the n scan window sizes defined in step S305, and the process proceeds to step S307.

  In the scan position changing step S307, the position of the scan window for raster scanning is defined in the detection region defined in step S303, and the process proceeds to step S308. In the object detection step S308, a still image object detection process is performed on the scan position defined in step S307 within the detection area defined in step S303 using the scan window size defined in step S306, and the process proceeds to step S309. move on.

  In the detection area scan completion determination step S309, it is determined whether the entire detection area defined in step S303 has been scanned using the scan window size defined in step S306. If the scan is completed, the process proceeds to step S310. If the scan is not completed, the process proceeds to step S307 to change the scan position.

  In step S310, the use of all scan window sizes is completed, it is determined whether the detection area defined in step S303 has been scanned using the n scan window sizes defined in step S305. If the scan is completed, the process proceeds to step S311. If the scan is not completed, the process proceeds to step S306, and the scan window for performing the still image object detection process next from the n scan window sizes selected in step S305. Select a size.

  In the entire area scan completion determination step S311, it is determined whether or not the entire image input in step S302 has been scanned. If the scan is completed, the process proceeds to step S312, and if not completed, the process proceeds to step S303, and the detection area is changed.

  The detection frame display step S312 displays all the object detection frames obtained in step S308, and proceeds to step S302. Each step may be described by a program that can be executed by a computer.

  FIG. 4 is a flowchart showing control of the detection frame size histogram information acquisition apparatus according to Embodiment 1 of the present invention.

  In parameter setting step S401, the number and size of areas into which the input image is divided, the start scale of the still image object detection process, the scale magnification, the number of times of resizing, etc. are defined, and the process proceeds to step S402. An image input step S402 inputs an image. At this time, the input image is divided into regions defined in step S401, and the process proceeds to step S403. In scan window size changing step S403, the size of the scan window for raster scanning the entire screen is defined, and the flow proceeds to step S404. In the scan position changing step S404, the position of the scan window for raster scanning is defined, and the process proceeds to step S405. In the object detection step S405, the still image object detection process is performed at the position specified in step S404 using the scan window size specified in step S403. If an object is detected, the process proceeds to step S406. If an object is not detected, the process proceeds to step S408.

  In the detection frame size S acquisition / detection frame centroid position C calculation step S406, the size S of the object detection frame obtained in step S405 and the centroid position C of the object detection frame are calculated, and the process proceeds to step S407. In the voting step S407 for the detection frame size histogram, the size S of the object detection frame obtained in step S406 is voted for the detection frame size histogram of the region to which the barycentric position C of the object detection frame obtained in step S406 belongs. Proceed to

  In step S408, the entire screen scan completion determination step determines whether the entire screen has been scanned with the scan window size defined in step S403. If the entire screen has been scanned, the process proceeds to step S409. If not completed, the process proceeds to step S404, and the scan position is changed while maintaining the scan window size defined in step S403.

  In the designated resizing count elapse determination step S409, it is determined whether or not the scan window size has been resized the number of times set in step S401. If the specified number of times has been resized, the process proceeds to step S410. If not, the process proceeds to step S403, and the current scan window size is multiplied by the scale factor set in step S410 to change the scan window size.

  A learning completion determination step S410 determines whether or not voting (learning) for the detection frame size histogram of each region is sufficient. If the number of votes for the detection frame size histogram of each area is sufficient, the process proceeds to step S411. If not, the process proceeds to step S402, and an object is detected again from the input image, and the vote for the detection frame size histogram of each area is performed. Increase the number.

  The detection frame size histogram storage step S411 stores the detection frame size histogram of each region, its average value, and standard deviation. Each step may be described by a program that can be executed by a computer.

  Next, generation of detection frame size histogram information generated by the detection frame size histogram information acquisition apparatus described in FIG. 2 will be described with reference to FIG.

  FIG. 5 shows an explanatory diagram of the histogram according to the first embodiment of the present invention. This detection frame size histogram represents how easily an object having a detection frame size in each region in an image is easily obtained in a certain imaging environment.

  Next, still image object detection using the scan window size determining apparatus 101 according to Embodiment 1 of the present invention described in FIG. 1 will be described with reference to FIG. In FIG. 6, the upper body of a person is used as an object to be detected for still image object detection.

  The input image is divided into several areas (in this case, divided into 20 areas) by the image input unit 102 as indicated by dotted lines in FIG. 6, and an area number is assigned to each area. For each region, a predetermined number of detection frame sizes are obtained from each detection frame size histogram, and the number is assumed to be the detection frame size sampling number n, where n = 3. Then, detection frame size histogram information of each region is read from the detection frame size histogram information storage unit 103.

  Here, paying attention to area N0.8, area No. Suppose that 8 is the detection region of interest. The scan window size determining unit 104 selects the area number. For the eight detection frame size histograms, three scan window sizes are selected by weighted sampling based on the appearance probability of each detection frame size. Note that the three selected scan window sizes are S3, 1, S3, 2, S3, 3 in descending order.

  The area object detection unit 105 includes an area number. A still image object detection process is performed on a region 601 with a margin around 8 using three selected scan window sizes (S3, 1, S3, 2, S3, 3).

  Note that applying the still image object detection process while increasing the scan window size in the order of S3, 1, S3, 2, S3, 3 with respect to 601 means that the still image is reduced while reducing the 601 while keeping the scan window size constant. This is synonymous with the object detection processing, and FIG. 6 is drawn from the viewpoint of the latter.

  In this way, by using the prior information (detection frame size histogram) of the object detection frame size acquired in the imaging environment in advance, only the scan window size that is statistically easy to detect an object in each area in the input image is stopped. By performing the image object detection process, it is possible to expect high-speed processing and suppression of erroneous detection.

  FIG. 7 is an explanatory diagram of a scan window size selection step by weighted sampling according to Embodiment 1 of the present invention.

  Regarding the detection frame size histogram 701 of an arbitrary region read from the detection frame size histogram information storage unit 103, the horizontal axis is the detection frame size, the vertical axis is the appearance probability of each detection frame size, μ is the average value of the detection frame sizes, σ Represents the standard deviation of the detection frame size.

  Weighted sampling based on the appearance probability is performed on the scan window sizes (S1, S2, S3, S4) included in the range of μ ± σ, and n scan windows are selected.

  As a means for realizing weighted sampling, a roulette method can be mentioned. Reference numeral 702 represents a pie chart of the appearance probability of the scan window size (S1, S2, S3, S4) included in the range of μ ± σ. In the roulette method, the roulette is stopped by rotating the roulette on the pie chart. This is equivalent to selecting the scan window size of the selected area. That is, in the 702 pie chart, the larger the scan window size (the greater the appearance probability), the higher the probability of being selected by weighted sampling. In addition to the roulette method, for example, equidistant sampling, Markov chain Monte Carlo method (MCMC: MARKOV CHAIN MONTE CARLO), or the like may be used.

  FIG. 8 is an explanatory diagram of weighted sampling according to the shape of the detection frame size histogram according to Embodiment 1 of the present invention. A case where the detection frame size histogram of an arbitrary region read from the detection frame size histogram information storage unit 103 is convex is indicated by 801, and a case where the detection frame size histogram is flat is indicated by 802.

  In both 801 and 802, the horizontal axis represents the detection frame size, the vertical axis represents the appearance probability of each detection frame size, μ represents the average value of the detection frame sizes, and σ represents the standard deviation of the detection frame sizes. It is assumed that the detection frame size sampling number n from each detection frame size histogram is n = 3.

  When the detection frame size histogram is convex as shown in 801, the scan window size included in the range of μ ± σ is less than 3, or the scan window size selected by weighted sampling is less than 3. There is a case. In this case, even if there are three scan window sizes designated in advance, a scan window size of less than three finally selected is adopted.

  Further, when the detection frame size histogram is flat as indicated by 802, three scan window sizes selected by weighted sampling are employed.

  FIG. 9 is an explanatory diagram of the detection frame size histogram information acquisition apparatus according to Embodiment 1 of the present invention described in FIG. In FIG. 9, the upper body of a person is used as an object to be detected for still image object detection.

  The input image is divided into several areas (in this case, divided into 20 areas) by the image input unit 102 as indicated by dotted lines in FIG. 9, and an area number is assigned to each area. Reference numerals 901 and 902 denote object detection frames obtained by the screen object detection unit 202, and 903 and 904 denote the centers of gravity of the object detection frames.

  In this case, the area numbers to which 903 and 904 belong are respectively No. 8 and no. 13 Assuming that the object detection frame size of the object detection frame 901 is S8 and the detection frame size S13 of the detection frame 902, the detection frame size histogram voting unit 203 of each area is No. S8 bin of the detection frame size histogram 905 of the area represented by 8; Voting is performed on the bin of S13 in the detection frame size histogram 906 of the area represented by 13.

  After the processing of the screen object detection unit 202 and the detection frame size histogram voting unit 203 for each region is completed for all input images, the detection frame size histogram of each region, its average value, and standard deviation are stored.

  In this case, reference numeral 905 denotes an area number. 8 is easy to obtain an object having a scan window size S8. 13 indicates that an object having the scan window size S13 is easily obtained, and the detection frame size histogram indicates prior information on the detection frame size related to the imaging environment.

(Embodiment 2)
In the second embodiment of the present invention, a detection frame size histogram is used as in the first embodiment. Therefore, a configuration diagram, a control schematic diagram, and an explanatory diagram related to the detection frame size histogram information acquisition apparatus and the detection frame size histogram information acquisition method according to Embodiment 2 of the present invention are left to those described in Embodiment 1. .

  FIG. 10 is a block diagram of a false detection removal apparatus 1001 according to Embodiment 2 of the present invention.

In FIG. 10, the image input unit 102, the detection frame size histogram information storage unit 103, and the area object detection unit 105 are the same as those in FIG. 1, and the false detection removal unit 1002 is an object detection frame obtained by the area object detection unit 105. Is detected with reference to the detection frame size histogram information storage unit 103.

  The area object detection unit 105 performs still image object detection processing on the entire input image with a start scale, scale magnification, and number of resizings specified in advance. Reference numeral 1002 denotes a false detection removal unit that removes false detection frames from the object detection frames obtained by the region object detection unit 105 based on the detection frame size histogram information of each region.

  FIG. 11 is a flowchart of the false detection removal apparatus according to Embodiment 2 of the present invention. In parameter setting step S1101, the number and size of regions into which the input image is divided and the start scale, scale magnification, resize count, etc. of the still image object detection process are defined, and the process proceeds to step S1102.

  An image input step S1102 inputs an image. At this time, the input image is divided into regions defined in step S1101, and the process proceeds to step S1103. In scan window size changing step S1103, the size of the scan window for raster scanning the entire screen is defined, and the flow advances to step S1104.

  In scan position changing step S1104, the position of the scan window at the time of raster scanning is defined, and the process proceeds to step S1105. In the object detection step S1105, the still image object detection process is performed on the position defined in step S1104 using the scan window size defined in step S1103. If an object is detected, the process proceeds to step S1106, and if no object is detected, the process proceeds to step S1110.

  The detection frame size S acquisition / detection frame centroid position C calculation step S1106 calculates the size S of the object detection frame obtained in step S1105 and the centroid position C thereof, and the process proceeds to step S1107.

  The detection frame size histogram information reading step S1107 reads the detection frame size histogram information (average value μ, standard deviation σ) of the region to which the barycentric position C of the detection frame obtained in step S1106 belongs, and proceeds to step S1108.

  In the erroneous detection determination step S1108, it is determined whether or not the object detection frame size obtained in step S1105 is within a range of ± σ from the average value μ read in step S1107. If the object detection frame size obtained in step S1105 is not within the range of μ ± σ, the process proceeds to step S1109, and if it is, the process proceeds to step S1110.

  In the erroneous detection removal step S1109, the object detection frame determined as the erroneous detection frame in step S1108 is removed, and the process proceeds to step S1110. In step S1110, the entire screen scan completion determination step determines whether the entire screen has been scanned with the scan window size defined in step S1103. If the entire screen has been scanned, the process proceeds to step S1111. If not completed, the process proceeds to step S1104, and the scan position is changed while the scan window size defined in step S1103 is maintained.

The specified resizing count elapse determination step S1111 determines whether or not the scan window size has been resized the number of times set in step S1101. If it has been resized the specified number of times, the process proceeds to step S1112. If not, the process proceeds to step S1103, and the current scan window size is multiplied by the scale factor set in step S1101 to change the size of the scan window.
In the detection frame display step S1112, the object detection frame is displayed in a state where the erroneous detection frame determined to be erroneous detection in step S1108 is removed from the object detection frame obtained in step S1105. Each step may be described by a program that can be executed by a computer.

  FIG. 12 is a diagram for explaining the operation of the false detection removal apparatus according to Embodiment 2 of the present invention. The input image is divided into several areas (in this case, divided into 20 areas) by the image input unit 102 as indicated by dotted lines in FIG. 12, and an area number is assigned to each area. 1201 and 1202 are object detection frames obtained by the area object detection unit 105, respectively, and 1203 and 1204 represent their centroids.

  In this case, the area numbers to which 1203 and 1204 belong are respectively No. 9 and no. 13 Reference numerals 1205 and 1206 denote the area numbers read by the detection frame size histogram information storage unit 103, respectively. 9 and region no. 13 is a detection frame size histogram, and (μ9, σ9) and (μ13, σ13) are respectively region Nos. 9 and region no. 13 is an average value and standard deviation of 13 detection frame size histograms.

  Assuming that the object detection frame size of the object detection frame 1201 is S9 and the object detection frame size S13 of the object detection frame 1202, the erroneous detection removal unit 1002 determines whether or not S9 is within a range of μ9 ± σ9 as indicated by 1205, As shown at 1206, it is determined whether or not S13 is within the range of μ13 ± σ13.

  In this case, S9 is removed as an erroneous detection because it exists outside the range of μ9 ± σ9, and S13 is not removed because it exists within the range of μ13 ± σ13, and is displayed as an object detection frame. As described above, the object detection frame size that is difficult to statistically detect in each area in the input image by using the prior information (detection frame size histogram) of the object detection frame size acquired in the imaging environment in advance. The detection frame can be removed as a false detection.

  The scan window size determining device, the detection frame size histogram information acquiring device, and the scan window size determining method according to the present invention install an object detection device and divide an input image into regions when detecting an object. By determining the range of the detection frame size, it has an effect of efficiently detecting an object and eliminating false detection, and can be applied to a monitoring system or the like.

DESCRIPTION OF SYMBOLS 101 Scan window size determination apparatus 102 Image input part 103 Detection frame size histogram information storage part 104 Scan window size determination part 105 Area | region object detection part

Claims (7)

A detection frame size histogram information storage unit that holds statistical information of a detection frame size for each of the plurality of regions of an image divided into a plurality of regions;
A scan window size determining unit that determines a scan window size of the region based on the statistical information stored in the detection frame size histogram information storage unit;
A scan window size determining apparatus comprising: An image input unit for inputting an image and an object detection unit;
The image input unit divides the input image into the plurality of regions,
2. The object detection unit according to claim 1, wherein the object detection unit detects an object for each of the plurality of regions using a scan window size determined for each of the plurality of regions by the scan window size determination unit. Scan window size determination device.   The scan window size determining apparatus according to claim 1, wherein the detection frame size histogram information is an average value and a variance of detection frame sizes when an object is detected for each region.   4. The scan window size determining apparatus according to claim 1, wherein the scan window size determining unit determines a scan window size using weighted sampling from the statistical information. A detection frame size histogram information acquisition device for generating statistical information of a detection frame size stored in the detection frame size histogram information storage unit according to claim 1,
An image input unit that divides the input image into a plurality of regions;
An object detection unit for detecting an object of the input image using a plurality of scan window sizes;
Detection frame size histogram voting for voting the detection frame size of the detection frame with respect to the detection frame size histogram of the region where the center of gravity of the obtained object detection frame is located when an object is detected by the object detection unit And a detection frame size histogram information acquisition device. A detection frame size histogram information storage unit that holds statistical information of a detection frame size for each of the regions divided into a plurality of regions;
An object detection unit for detecting an object of the input image using a plurality of scan window sizes;
When the object detection unit detects an object, using the detection frame size histogram of the region where the center of gravity of the obtained object detection frame is located, it is determined whether or not the object detection frame is erroneously detected. And
A scan window sizing device. A scan window size determining method for determining a scan window size using a computer,
A detection frame size histogram information storage step for holding statistical information of the detection frame size for each of the regions divided into a plurality of regions;
A scan window size determining step for determining a scan window size of the region based on the statistical information stored in the detection frame size histogram information storing step;
A method for determining a scan window size.