JP2014142760A - Image detection device, control program and image detection method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a technology for reducing throughput about a processing object image as the object of the detection of a detection object image.SOLUTION: A detection part 3 performs detection processing of detecting a region which is highly likely to be a detection object image having the same size as that of a detection frame as a detection result region to a pickup image. The detection part 3 associates a plurality of types of detection frames with a plurality of pickup images in a distributed manner such that at least one type of detection frame included in the plurality of types of detection frames having different sizes is associated with each of the plurality of pickup images including a processing object image. The detection part 3 performs detection processing to each of the plurality of pickup images by using at least one type of detection frame associated with each pickup image. A detection object image specification part 7 specifies the detection object image in the processing object image on the basis of the detection result region detected about the plurality of pickup images in the detection part 3.

Description

  The present invention relates to a technique for detecting a detection target image from a processing target image.

  Patent Documents 1 to 3 disclose techniques for detecting a detection target image from a processing target image.

JP 2008-27058 A JP 2009-175821 A JP 2011-221791 A

  Now, when detecting a detection target image from a processing target image, reduction of the processing amount about a processing target image is desired.

  Therefore, the present invention has been made in view of the above-described points, and an object thereof is to provide a technique capable of reducing the processing amount of a processing target image to be detected. To do.

  In order to solve the above problems, one aspect of an image detection device according to the present invention is an image detection device that detects a detection target image from a processing target image, and uses the detection frame to detect the detected image. A detection unit that performs a detection process of detecting an area that is likely to be the detection target image having the same size as a frame as a detection result area, and a detection target image specifying unit that specifies the detection target image in the processing target image The detection unit includes at least one type of detection included in a plurality of types of detection frames having different sizes with respect to each of a plurality of captured images including the processing target image captured at different timings. The plurality of types of detection frames are dispersed and associated with the plurality of captured images so that the frames correspond to each other, and each of the plurality of captured images corresponds to the captured image. The detection target image specifying unit performs the detection process using at least one type of detection frame, and the detection target image specifying unit detects the processing target image based on the detection result region detected for the plurality of captured images by the detection unit. In step (b), the detection target image is specified.

  In the image detection apparatus according to the aspect of the invention, the plurality of types of detection frames include a reference detection frame having a reference size and a non-reference detection frame having a size different from the reference size, and the detection unit includes When performing the detection process on the captured image using the non-reference detection frame, the non-reference detection frame is resized so that the size matches the reference size, and the non-reference detection frame is resized. The size of the captured image is changed according to the size, and the size change detection frame that is the non-reference detection frame after the size change is moved with respect to the size changed image that is the captured image after the size change. It is determined whether there is a high possibility that the image within the size change detection frame in the changed image is the detection target image.

  In one aspect of the image detection apparatus according to the present invention, the detection unit uses detection frames included in the plurality of types of detection frames for the captured images other than the processing target image in the plurality of captured images. As a result of performing the detection process, if there is a region in the captured image that is very likely to be the detection target image having the same size as the detection frame, the detection frame is also detected for the processing target image. The detection process is performed using.

  In the image detection device according to the aspect of the invention, the detection target image specifying unit may be based on a detection result frame that is an outer frame of the detection result area detected for the plurality of captured images in the detection unit. The detection target image is specified in the processing target image, the detection target image specifying unit divides the processing target image on which the detection result frame is overlaid into a plurality of blocks, and the detection target image specifying unit is When the number of captured images in which the detection result frame overlapping the block is obtained among the plurality of captured images is equal to or less than a threshold value for the block overlapping the detection result frame in the plurality of blocks Specifies the detection target image in the processing target image without using the detection result frame overlapping the block.

  In the image detection apparatus according to the aspect of the invention, the detection target image specifying unit may overlap the block of the plurality of captured images with respect to the block that overlaps the detection result frame in the plurality of blocks. Even if the number of captured images from which the detection result frame is obtained is equal to or less than a threshold value, the detection result image is in the detection result region having the detection result frame overlapping the block as an outer frame. When the detection result region having a very high possibility is included, the detection target image is specified in the processing target image using the detection result frame overlapping the block.

  In the image detection device according to the aspect of the invention, the detection target image specifying unit may detect the detection result region of the detection result region detected for the plurality of captured images in the detection unit. The detection target image is specified in the processing target image based on a detection accuracy value indicating the likelihood of being the target image.

  In the aspect of the image detection apparatus according to the present invention, the detection target image specifying unit weights the detection accuracy value for the detection result region detected for the plurality of captured images in the detection unit. The detection target image is specified in the processing target image based on the detection accuracy value, and the detection target image specifying unit is configured to detect the detection accuracy value of the detection result area detected for the captured image. When the weighting is performed on the detection accuracy value, the weighting of the detection accuracy value is reduced as the imaging timing of the captured image is further away from the imaging timing of the processing target image.

  In one aspect of the image detection apparatus according to the present invention, the detection target image is a human face image.

  An aspect of the control program according to the present invention is a control program for controlling an image detection apparatus that detects a detection target image from a processing target image. The image detection apparatus uses (a) a detection frame. A step of performing a detection process on the captured image to detect, as a detection result area, an area that is highly likely to be the detection target image having the same size as the detection frame; and (b) the detection in the processing target image. A plurality of types having different sizes with respect to each of a plurality of captured images including the processing target image captured at different timings in the step (a). The plurality of types of detection frames are distributed and associated with the plurality of captured images so that at least one type of detection frame included in the detection frames corresponds to the plurality of detection frames. For each of the images, the detection processing is performed using at least one type of detection frame corresponding to the captured image, and in the step (b), the plurality of captured images detected in the step (a) is detected. Based on the detection result area, the detection target image is operated to specify the detection target image in the processing target image.

  An aspect of the image detection method according to the present invention is an image detection method for detecting a detection target image from a processing target image. (A) Using a detection frame, A step of performing a detection process of detecting a region having a high possibility of being the detection target image of the same size as a detection result region, and (b) a step of specifying the detection target image in the processing target image, In (a), for each of a plurality of captured images including the processing target image captured at different timings, at least one type of detection frame is included in a plurality of types of detection frames having different sizes. In order to correspond, the plurality of types of detection frames are distributed and associated with the plurality of captured images, and each of the plurality of captured images is at least corresponding to the captured image. The detection process is performed using one type of detection frame, and in the process target image based on the detection result area detected for the plurality of captured images in the process (a) in the process (b). The detection target image is specified.

  According to the present invention, it is possible to reduce the amount of processing for a processing target image.

It is a figure which shows the structure of an image detection apparatus. It is a figure which shows the structure of the several functional block with which an image detection apparatus is provided. It is a figure which shows a mode that several types of detection frames are matched with the some use captured image. It is a flowchart which shows operation | movement of a detection part. It is a figure which shows the structure of a detection part. It is a figure for demonstrating operation | movement of a detection part. It is a figure for demonstrating operation | movement of a detection part. It is a figure for demonstrating operation | movement of a detection part. It is a figure for demonstrating operation | movement of a detection part. It is a figure for demonstrating operation | movement of a detection part. It is a figure for demonstrating operation | movement of a detection part. It is a figure which overlaps and shows a detection result frame on a process target image. It is a figure for demonstrating the production | generation method of an output value map. It is a figure for demonstrating the production | generation method of an output value map. It is a figure which shows an output value map. It is a figure for demonstrating the search method of a local maximum point. It is a figure for demonstrating the search method of a local maximum point. It is a figure which shows distribution of the detection accuracy value vicinity of the maximum point in an output value map. It is a figure for demonstrating the determination method of a detection image area | region. It is a flowchart which shows operation | movement of a detection target image determination part. It is a figure which overlaps and shows a detection image area | region and the detection result frame after integration on a process target image. It is a figure which shows a mode that the process target image with which the detection result frame was overlaid is divided | segmented into several blocks.

  FIG. 1 is a diagram illustrating a configuration of an image detection apparatus 1 according to an embodiment. The image detection apparatus 1 according to the present embodiment detects a detection target image from a captured image indicated by input image data. The image detection apparatus 1 is used, for example, in a surveillance camera system, a digital camera system, or the like. In the present embodiment, the detection target image is, for example, a human face image. Hereinafter, simply speaking “face image” means a human face image. A captured image to be detected from the detection target image is referred to as a “processing target image”. The image detection apparatus 1 according to the present embodiment assumes that a plurality of captured images having different imaging timings, including processing target images, but having different imaging timings indicate the same image, from the processing target images. When detecting a plurality of captured images. The detection target image in the image detection apparatus 1 may be an image other than a face image.

  As shown in FIG. 1, the image detection apparatus 1 includes a CPU (Central Processing Unit) 10 and a storage unit 11. The storage unit 11 includes a ROM (Read Only Memory), a RAM (Random Access Memory), and the like. The storage unit 11 stores a control program 12 for controlling the operation of the image detection apparatus 1 and the like. Various functions of the image detection apparatus 1 are realized by the CPU 10 executing the control program 12 in the storage unit 11. In the image detection apparatus 1, a plurality of functional blocks as shown in FIG. 2 are formed by executing the control program 12.

  As shown in FIG. 2, the image detection apparatus 1 includes an image input unit 2, a detection unit 3, and a detection target image specifying unit 7 as functional blocks. Hereinafter, after describing the schematic operation of each functional block included in the image detection apparatus 1, the detailed operation of each functional block will be described. Various functions provided in the image detection apparatus 1 may be realized by hardware circuits instead of function blocks.

<Overview of operation of image detection apparatus>
The image input unit 2 is sequentially input with a plurality of pieces of image data respectively indicating a plurality of captured images sequentially captured by an imaging unit (camera) included in the surveillance camera system or the like. The image input unit 2 outputs a plurality of image data respectively indicating K (K ≧ 2) captured images including the processing target image, which are used when the detection target image is detected from the processing target image. Hereinafter, the K captured images may be collectively referred to as a “used captured image group”. In addition, each of a plurality of (K) captured images constituting the used captured image group may be referred to as a “used captured image”. The plurality of used captured images constituting the used captured image group are captured at different timings.

  The image input unit 2 may use each captured image obtained by the imaging unit as a processing target image, or may use a captured image obtained every few seconds among the captured images obtained by the imaging unit as a processing target image. In the imaging unit, for example, L (L ≧ 2) captured images are captured per second. That is, the imaging frame rate at the imaging unit is Lfps (frame per second). For example, L = 30 is set.

  In the captured image obtained by the imaging unit, M (M ≧ 2) pixels are arranged in the row direction, and N (N ≧ 2) pixels are arranged in the column direction. The resolution of the captured image obtained by the imaging unit is, for example, VGA (Video Graphics Array), and M = 640 and N = 480.

  Hereinafter, the size of an area in which m (m ≧ 1) pixels are arranged in the row direction and n (n ≧ 1) pixels are arranged in the column direction is represented by mp × np (p is a meaning of a pixel). In addition, among a plurality of values arranged in a matrix, a value located in the m-th row and the n-th column with reference to the upper left may be referred to as an m × n-th value.

  The detection unit 3 uses a plurality of pieces of image data each indicating a plurality of use captured images constituting the use captured image group output from the image input unit 2, and for each of the plurality of use captured images. An area that is highly likely to be a face image in the use captured image is detected.

  The detection target image specifying unit 7 includes a map generation unit 4, a maximum point search unit 5, and a detection target image determination unit 6. The detection target image specifying unit 7 specifies a face image in the processing target image based on the detection result of the detection unit 3.

  The map generation unit 4 generates an output value map indicating the distribution in the processing target image with respect to the detection accuracy value indicating the likelihood as the face image based on the detection result of the detection unit 3. The maximum point search unit 5 searches for a maximum point of the detection accuracy value in the output value map generated by the map generation unit 4. The detection target image determination unit 6 determines that a predetermined area including a pixel at the same position as the maximum point obtained by the maximum point search unit 5 in the processing target image is a face image. Thereby, the detection target image specifying unit 7 specifies a face image in the processing target image. As a result, the image detection apparatus 1 detects a face image from the processing target image.

<Detailed operation description of image detection apparatus>
<Detection process>
The detection unit 3 uses the detection frame to perform a detection process of detecting, as a detection result region, a region that is highly likely to be a face image having the same size as the detection frame in the used captured image. Hereinafter, simply speaking “detection process” means this detection process in the detection unit 3. The detection process will be described in detail later.

  The detection unit 3 uses a plurality of types of detection frames having different sizes in order to detect face images of various sizes in the processing target image. The detection unit 3 is configured so that at least one type of detection frame included in the T types (T ≧ 2) of detection frames corresponds to each of the K use captured images constituting the use captured image group. A plurality of types of detection frames are dispersed and associated with the K used captured images. Then, the detection unit 3 performs detection processing for each of the K used captured images constituting the used captured image group using at least one type of detection frame corresponding to the used captured image.

  FIG. 3 is a diagram illustrating a state in which a plurality of types of detection frames 100 used in the detection unit 3 are distributed and associated with a plurality of used captured images that form a used captured image group. In the present embodiment, for example, K = 3, and the used captured image group includes three captured images that are continuously captured by the imaging unit. Assuming that the frame rate at the imaging unit is 30 fps, the imaging interval of the three used captured images constituting the used captured image group is (1/30) seconds. In the present embodiment, among the three used captured images constituting the used captured image, for example, the last captured captured image is used as the processing target image.

  In FIG. 3, a plurality of types of detection frames 100 are distributed and associated with the used captured images of the (k-2) th frame, the (k-1) th frame, and the kth frame constituting the used captured image group. It is shown that it is being done. In the example of FIG. 3, the use captured image of the kth frame is the processing target image. Hereinafter, the operation of the image detection apparatus 1 will be described by taking as an example a case where the use captured image group is configured by the (k-2) th frame, the (k-1) th frame, and the kth frame.

  As shown in FIG. 3, a plurality of types of detection frames 100 having different sizes from each other are dispersed and used for the use imaged images of the (k−2) th frame, the (k−1) th frame, and the kth frame. It is associated. In the example of FIG. 3, the detection frames 100 are used one by one in order from the smallest size, the (k−2) th frame used captured image, the (k−1) th frame used captured image, and the kth frame used. The captured images are associated with the used captured images in the order in which the order of the captured images is repeated. That is, the (3 × s + 1) th (s = 0, 1, 2,...) Small detection frame 100 is associated with the used captured image of the (k−2) th frame, and (k−1). The (3 × s + 2) th smallest detection frame 100 is associated with the used captured image of the frame, and the (3 × s + 3) th smallest detection frame 100 is associated with the used captured image of the kth frame. It is attached.

  Thus, in the present embodiment, for each of the three used captured images constituting the used captured image group, the detection frame 100 is not overlapped between the three used captured images. Every two items are associated in ascending order of size. In the present embodiment, a plurality of types of detection frames 100 are associated with each used captured image. For example, when T = 30, ten types of detection frames 100 are associated with each used captured image.

  When the captured image of the (k-1) th frame is the processing target image, the used captured image group includes the (k-3) th frame, the (k-2) th frame, and the (k-1). It is composed of the captured image of the frame. In this case, the same detection frame 100 as the detection frame 100 associated with the kth frame is associated with the captured image of the (k-3) th frame.

  When the captured image of the (k-2) th frame is the processing target image, the used captured image group includes the (k-4) th frame, the (k-3) th frame, and (k-2). It is composed of the captured image of the frame. In this case, the same detection frame 100 as the detection frame 100 associated with the (k-1) frame is associated with the captured image of the (k-4) frame.

  When the captured image of the (k-3) th frame is the processing target image, the used captured image group includes the (k-5) th frame, the (k-4) th frame, and the (k-3). It is composed of the captured image of the frame. In this case, the same detection frame 100 as the detection frame 100 associated with the (k-2) frame is associated with the captured image of the (k-5) frame.

  To generalize the above points, when the captured image group includes the captured image of the (k + 3t) frame (t is an integer excluding zero), and the detection frame 100 is associated with the captured image, The same detection frame 100 as the k-th detection frame 100 is associated. When the captured image group includes the captured image of the (k + 3t−1) frame and the detection frame 100 is associated with the captured image, the detection frame 100 of the (k−1) frame is The same detection frame 100 is associated. When the captured image group includes the captured image of the (k + 3t-2) frame, and the detection frame 100 is associated with the captured image, the detection frame 100 of the (k-2) frame is The same detection frame 100 is associated.

  FIG. 4 illustrates the detection unit 3 when the detection unit 3 performs detection processing using a plurality of types of detection frames corresponding to the use captured image for each of the plurality of use captured images constituting the use captured image group. It is a flowchart which shows a series of operation | movement.

  As shown in FIG. 4, when image data indicating a use captured image of the (k−2) th frame is input from the image input unit 2 in step s1, the detection unit 3 receives the image in step s2. Using the data, detection processing is performed on the used captured image of the (k-2) frame using each detection frame corresponding to the used captured image of the (k-2) frame. That is, for each detection frame corresponding to the (k-2) -th frame used captured image, the detection unit 3 uses the detection target image having the same size as the detection frame in the (k-2) -frame used captured image. Detection processing is performed to detect a region having a high possibility as a detection result region.

  After execution of step s2, in step s3, when the image data indicating the used captured image of the (k−1) th frame is input from the image input unit 2, the detection unit 3 receives the image data in step s4. Using the detection frames corresponding to the used captured image of the (k-1) th frame, the detection processing is performed on the used captured image of the (k-1) th frame.

  After execution of step s4, in step s5, when the image data indicating the use-captured image of the k-th frame is input from the image input unit 2, the detection unit 3 uses the image data in step s6 to obtain k. Using each detection frame corresponding to the used captured image of the frame, detection processing is performed on the used captured image of the kth frame.

  As described above, in the present embodiment, the T types of detection frames are distributed and associated with the K used captured images including the processing target images that constitute the used captured image group. Fewer types of detection frames are associated with the target image than the T types of detection frames. Therefore, for the processing target image, detection processing is performed using fewer types of detection frames than the T types of detection frames. Therefore, the processing amount for the processing target image can be reduced as compared with the case where the detection processing is performed using T types of detection frames for the processing target image.

<Details of detection processing>
Next, details of the detection process will be described. FIG. 5 is a diagram illustrating a configuration of the detection unit 3. As shown in FIG. 5, the detection unit 3 includes an image cropping unit 30, a feature amount extraction unit 31, a discriminator 32, a determination unit 33, and an image size change unit 34.

  In the present embodiment, as will be described later, the feature amount extraction unit 31 extracts a feature amount from an input image. The feature amount extraction unit 31 needs to input an image of a reference size in order to extract the feature amount from the input image.

  On the other hand, in this embodiment, T types of detection frames having different sizes include a detection frame having the same size as the reference size and a detection frame having a size different from the reference size. Hereinafter, a detection frame having the same size as the reference size is referred to as a “reference detection frame”, and a detection frame having a size different from the reference size is referred to as a “non-reference detection frame”. In the present embodiment, the detection frame having the smallest size among the T types of detection frames is the reference detection frame. Therefore, (T-1) types of detection frames among the T types of detection frames are non-reference detection frames. The size of each of the (T-1) types of non-reference detection frames is larger than the reference size. The size of the reference detection frame is, for example, 16p × 16p. The (T-1) types of non-reference detection frames include, for example, a non-reference detection frame having a size of 18p × 18p and a non-reference detection frame having a size of 20p × 20p.

  In the present embodiment, when performing detection processing using a reference detection frame for a used captured image, the detection unit 3 moves the reference detection frame with respect to the used captured image while moving the reference detection frame within the reference detection frame. A face image is detected for the image to determine whether the image is highly likely to be a face image. And the detection part 3 makes the area | region (image in a reference | standard detection frame) determined with the possibility that it is a face image high in a use captured image as a detection result area | region.

  On the other hand, when performing detection processing using the non-reference detection frame for the used captured image, the detection unit 3 changes the size of the non-reference detection frame so that the size matches the reference size. Then, the detection unit 3 changes the size of the used captured image in accordance with the size change of the non-reference detection frame. The detection unit 3 detects the face image for the image in the non-reference detection frame while moving the non-reference detection frame whose size has been changed with respect to the used captured image whose size has been changed, It is determined whether or not the image is likely to be a face image. Then, the detection unit 3 performs the size change based on the region (the image in the non-reference detection frame after the size change) that is determined to be highly likely to be a face image in the use-captured image after the size change. A region that has a high possibility of being a face image in a used captured image of an original size that is not known is identified, and the region is set as a detection result region.

  Hereinafter, the used captured image after the size change when the non-reference detection frame is used for the used captured image and the detection process is performed is referred to as a “size-changed image”. In addition, the non-reference detection frame after the size change when the detection process is performed using the non-reference detection frame for the used captured image is referred to as a “size change detection frame”.

  As described above, in the present embodiment, the operation of the detection unit 3 when the detection unit 3 performs the detection process on the used captured image using the reference detection frame, and the detection unit 3 operates on the used captured image. Thus, the operation of the detection unit 3 when performing the detection process using the non-reference detection frame is different. The operation of the detection unit 3 will be described in detail below.

  In the detection unit 3, when the reference detection frame is used for the detection process, the image cutout unit 30 sets a reference detection frame for the use captured image, and the image within the reference detection frame from the use captured image. Is input to the feature amount extraction unit 31. On the other hand, when the non-reference detection frame is used for the detection process, the image cropping unit 30 applies the size change image obtained by resizing the use captured image by the image size changing unit 34 to the size change image. A size change detection frame obtained by resizing the non-reference detection frame is set, and an image within the size change detection frame is cut out from the size change image and input to the feature amount extraction unit 31.

  Here, since the size of the reference detection frame coincides with the reference size, the size of the image in the reference detection frame cut out by the image cutout unit 30 becomes the reference size. In addition, since the size of the size change detection frame matches the reference size, the size of the image in the size change detection frame cut out by the image cutout unit 30 becomes the reference size. Therefore, an image having a reference size is always input to the feature amount extraction unit 31.

  The feature amount extraction unit 31 uses, for example, a Haar-like feature amount or LBP (Local Binary Pattern) from the input image (the image in the reference detection frame in the used captured image or the image in the size change detection frame in the size change image). Extract feature quantities such as feature quantities.

  The discriminator 32 performs face detection on the image cut out by the image cutout unit 30 based on the feature amount extracted by the feature amount extraction unit 31 and the learning data, and as a result, the image is a face image. The detection accuracy value indicating the probability is output as a real value. That is, it can be said that the detection accuracy value output as an output value from the discriminator 32 indicates the face image-likeness (face-likeness) of the image in the reference detection frame or the image in the size change detection frame. As the discriminator 32, for example, SVM (Support Vector Machine) or Adaboost is used.

  If the detection accuracy value output from the discriminator 22 is equal to or greater than the threshold value, the determination unit 33 determines that there is a high possibility that the image cut out by the image cutout unit 30 is a face image. That is, when the reference detection frame is used, the determination unit 33 determines that the image in the reference detection frame in the use target image is a region that is highly likely to be a face image having the same size as the reference detection frame. To do. Further, when the non-reference detection frame is used, the determination unit 33 is an area in which there is a high possibility that the image in the size change detection frame in the size change image is a face image having the same size as the size change detection frame. Judge that there is.

  On the other hand, if the detection accuracy value output from the discriminator 22 is less than the threshold, the determination unit 33 determines that there is a high possibility that the image cut out by the image cutout unit 30 is not a face image. That is, when the reference detection frame is used, the determination unit 33 determines that the image in the reference detection frame in the used captured image is not a region that is highly likely to be a face image having the same size as the reference detection frame. To do. When the non-reference detection frame is used, the determination unit 33 determines that the image in the size change detection frame in the size change image is likely to be a face image having the same size as the size change detection frame. Judge that there is no.

  When the determination unit 33 determines that the image in the reference detection frame in the use target image is a region that is highly likely to be a face image having the same size as the reference detection frame, the image is used as a detection result region, and the reference detection is performed. Let the frame be the detection result frame.

  Further, when the determination unit 33 determines that the image in the size change detection frame in the size change image is a region having a high possibility of being a face image having the same size as the size change detection frame, the determination unit 33 is the outline frame of the region. The size change frame is set as a temporary detection result frame. Then, the determination unit 33 identifies a region that is highly likely to be a face image having the same size as the non-reference detection frame in the use captured image of the original size before the size change based on the temporary detection result frame. The area is set as a detection result area, and the outer frame of the detection result area is set as a final detection result frame.

<Detection process using reference detection frame>
Next, the detection unit 3 moves the reference detection frame with respect to the used captured image, and determines whether the image in the reference detection frame is likely to be a face image. The operation will be described. 6 to 9 are diagrams for explaining the operation of the detection unit 3. The detection unit 3 detects a face image with respect to an image in the reference detection frame while raster scanning the reference detection frame. In this embodiment, the reference detection frame is a reference frame of the minimum size, and the reference frame of the minimum size is associated with the used captured image of the (k-2) th frame as shown in FIG. Therefore, when the reference detection frame is used, the detection process is performed on the used captured image of the (k-2) th frame.

  As illustrated in FIG. 6, the image cropping unit 30 first sets the reference detection frame 100 on the upper left of the used captured image 20 (used captured image of the (k−2) frame) associated with the reference detection frame 100. Then, the image in the reference detection frame 100 is cut out. Thereafter, the feature quantity extraction unit 31 extracts the feature quantity from the image cut out by the image cutout unit 30. Since the size of the reference detection frame matches the reference size, an image of the reference size is input to the feature amount extraction unit 31.

  The discriminator 32 obtains a detection accuracy value for the image cut out by the image cutout unit 30 based on the feature amount extracted by the feature amount extraction unit 31 and the learning data. When the detection accuracy value obtained by the discriminator 32 is equal to or greater than the threshold value, the determination unit 33 determines that the image cut by the image cutout unit 30, that is, the upper left reference detection frame 100 in the used captured image 20. It is determined that there is a high possibility that the area is a face image, the area is set as a detection result area, and the reference detection frame 100 which is an outer frame of the area is set as a detection result frame.

  Next, the image cropping unit 30 moves the reference detection frame 100 slightly to the right in the used captured image 20. For example, the image cutout unit 30 moves the reference detection frame 100 to the right by one pixel or several pixels. Then, the image cutout unit 30 cuts out the image in the reference detection frame 100 after movement in the used captured image 20.

  Thereafter, the feature amount extraction unit 31 extracts a feature amount from the image cut out by the image cutout unit 30, and the discriminator 32 uses the feature cutout and the learning data for the image cut out by the image cutout unit 30. Find the detection accuracy value. When the detection accuracy value obtained by the discriminator 32 is equal to or greater than the threshold value, the determination unit 33 determines that the image cut out by the image cutout unit 30 is highly likely to be a face image, and The image is used as a detection result area, and the reference detection frame 100 set by the image cropping unit 30 that is the outer frame of the image is used as the detection result frame.

  Thereafter, the detection unit 3 operates in the same manner, and as illustrated in FIG. 7, when the reference detection frame 100 moves to the right end of the use captured image 20, the detection unit 3 detects the image in the reference detection frame 100 at the right end. The detection accuracy value is obtained. If the obtained detection accuracy value is equal to or greater than the threshold value, the detection unit 3 sets the image in the rightmost reference detection frame 100 as a detection result region, and uses the rightmost reference detection frame 100 as a detection result frame. To do.

  Next, as illustrated in FIG. 8, the image cutout unit 30 cuts the image in the reference detection frame 100 after moving the reference detection frame 100 to the left end of the used captured image 20 while slightly lowering the reference detection frame 100. The image cutout unit 30 moves the reference detection frame 100 downward by, for example, one pixel or several pixels in the vertical direction (column direction). Thereafter, the feature amount extraction unit 31 extracts the feature amount from the image cut out by the image cutout unit 30, and the discriminator 32 uses the feature cutout and the learning data for the image cut out by the image cutout unit 30. Find and output the detection accuracy value. When the detection accuracy value output from the discriminator 32 is equal to or greater than the threshold value, the determination unit 33 determines that the image cut out by the image cutout unit 30 is highly likely to be a face image, and The image is set as a detection result area, and the reference detection frame 100 set by the image cutout unit 30 is set as a detection result frame.

  Thereafter, the detection unit 3 operates in the same manner, and as shown in FIG. 9, when the reference detection frame 100 moves to the lower right of the use captured image 20, the detection unit 3 moves within the reference detection frame 100 on the lower right. The detection accuracy value for the image is obtained. If the obtained detection accuracy value is equal to or greater than the threshold value, the detection unit 3 sets the image in the lower right reference detection frame 100 as a detection result region and uses the lower right reference detection frame as the detection result frame. And

  As described above, the detection unit 3 uses the reference detection frame, and in the use captured image associated with the reference detection frame, the region that is highly likely to be a face image having the same size as the reference detection frame. Is detected as a detection result area.

<Detection process using non-reference detection frame>
When the detection unit 3 performs the detection process using the non-reference detection frame, the image cropping unit 30 is configured so that the size of the non-reference detection frame matches the reference size (the size of the reference detection frame). Resize the non-reference detection frame. Then, the image size changing unit 34 changes the size of the used captured image associated with the non-reference detection frame by the same size change ratio as the non-reference detection frame.

  In the present embodiment, since the reference size is 16p × 16p, for example, when a non-reference detection frame having a size of Rp × Rp (R> 16) is used, the image cropping unit 30 performs the non-reference detection. The non-reference detection frame is reduced by multiplying the vertical width (vertical width) and horizontal width (horizontal width) of the detection frame by (16 / R), respectively, and a size change detection frame is generated. Then, the image size changing unit 34 multiplies (16 / R) the vertical width (number of pixels) and the horizontal width (number of pixels) of the used captured image associated with the non-reference detection frame, respectively. Reduce and generate a resized image. After that, the detection unit 3 moves the size change detection frame with respect to the size change image and moves the image in the size change detection frame to the size in the same manner as the processing described with reference to FIGS. It is determined whether there is a high possibility that the face image has the same size as the change detection frame. That is, the detection unit 3 performs processing for detecting an area that is highly likely to be a face image having the same size as the size change detection frame in the size change image using the size change detection frame. Hereinafter, this process is referred to as a “size-changed version detection process”.

  In the size change version detection process, the detection unit 3 sets a size change detection frame for the size change image, and the image in the size change detection frame may be a face image having the same size as the size change detection frame. If it is determined that the property is high, the size change detection frame that is the outer frame of the image is set as a temporary detection result frame. As a result, the detection unit 3 obtains at least one temporary detection result frame for the resized image. Since the size of the size change detection frame matches the reference size, an image that matches the reference size is input to the feature amount extraction unit 31 in the size change version detection process.

  When the detection unit 3 obtains at least one temporary detection result frame for the size-changed image, the determination unit 33 converts the at least one temporary detection result frame into a detection result frame according to the captured image of the original size. Convert.

  Specifically, the determination unit 33 first sets at least one obtained temporary detection result frame for the resized image. FIG. 10 is a diagram illustrating a state in which the temporary detection result frame 130 is set for the size-changed image 120. In the example of FIG. 10, a plurality of temporary detection result frames 130 are set for the resized image 120.

  Next, as illustrated in FIG. 11, the determination unit 33 uses the resized image 120 by enlarging (resizing) the resized image 120 in which the temporary detection result frame 130 is set and returning the resized image to the original size. The captured image 20 is converted. As a result, the temporary detection result frame 130 set in the size-changed image 120 is also enlarged, and the temporary detection result frame 130 is converted into a detection result frame 150 corresponding to the used captured image 20, as shown in FIG. The A region within the detection result frame 150 in the use captured image 20 is a detection result region that is highly likely to be a face image having the same size as the non-reference detection frame in the use captured image 20. Thereby, in the detection unit 3, based on the temporary detection result frame 130 obtained by the size-changed version detection process, a detection result region that is highly likely to be a face image having the same size as the non-reference detection frame in the used captured image. Identified.

  As described above, when the detection unit 3 performs the detection process on the used captured image using the non-reference detection frame, the non-reference detection frame whose size is changed to match the reference size, the non-reference detection frame The size-changed version detection process is performed using the used captured image whose size is changed in accordance with the size change of the reference detection frame. Thus, even when a detection frame having a size different from the reference size is used, an image of the reference size is input to the feature amount extraction unit 31. And the detection part 3 pinpoints the detection result area | region with high possibility that it is a face image of the same size as a non-reference | standard detection frame in a use captured image based on the result of a size-change version detection process.

  In addition, when performing the detection process on the used captured image using the non-reference detection frame, the detection unit 3 uses the non-reference detection frame and the used captured image as in the case of using the reference detection frame. It may be determined whether the image in the non-reference detection frame is likely to be a face image while moving the non-reference detection frame with respect to the used captured image without changing the size. In this case, each time the non-reference detection frame is set for the used captured image (each time the non-reference detection frame is moved), the image cropping unit 30 uses the size of the image in the non-reference detection frame as a reference. After changing the size, the image is input to the feature amount extraction unit 31. Therefore, every time a non-reference detection frame is set for a used captured image, an image size changing process is required. From the viewpoint of reducing the amount of processing, it is desirable to perform the processing after resizing the non-reference detection frame and the used captured image as in the above-described size-changed detection processing.

  In addition, when the size of the non-reference detection frame is an integral multiple of the size of the reference detection frame (reference size) in each of the vertical direction and the horizontal direction, the captured image used when generating the size-changed image Since each of the vertical width and the horizontal width can be (1 / integer) times, the used captured image can be accurately reduced using an averaging filter or the like. Therefore, it is desirable that the vertical width and the horizontal width of the non-reference detection frame are each an integral multiple of the vertical width and the horizontal width of the reference detection frame.

  In addition, for a plurality of types of detection frames associated with one used captured image, each set of two types of adjacent detection frames is arranged when the plurality of types of detection frames are arranged in order from the smallest size. In this case, it is desirable that the vertical width and horizontal width of the detection frame having the larger size are fixed integer multiples of the vertical width and horizontal width of the detection frame having the smaller size.

  For example, the reference detection frame and the first and second non-reference detection frames having different sizes are associated with the used captured image of the (k-2) th frame, and the reference detection frame, the first It is assumed that the size is smaller in the order of the non-reference detection frame and the second non-reference detection frame. In this case, for example, the vertical width and the horizontal width of the first non-reference detection frame are each set to be twice (in terms of area, 4 times) the vertical width and the horizontal width of the reference detection frame. Then, the vertical width and the horizontal width of the second non-reference detection frame are each set to be twice the vertical width and the horizontal width of the first non-reference detection frame.

  (K-2) When a plurality of types of detection frames of such a size are associated with the used captured image of the frame, the detection process is performed using the first non-reference detection frame. At this time, the vertical and horizontal widths of the used captured image of the (k−2) th frame are set to (½) times to generate a size-changed image for the first non-reference detection frame. When performing the detection process using the second non-reference detection frame, the size of the first non-reference detection frame is changed instead of reducing the use captured image of the (k-2) frame. The vertical and horizontal widths of the image are set to (1/2) times to generate a resized image for the second non-reference detection frame. As a result, by performing the reduction process with the same reduction ratio, it is possible to generate the resized images for the first and second non-reference detection frames.

  Also, the use captured image of the (k−1) th frame is associated with the third to fifth non-reference detection frames having different sizes, and the order of the third to fifth non-reference detection frames is the order. It is assumed that the size is small. In this case, the vertical width and the horizontal width of the third non-reference detection frame are set to, for example, 3 times (9 times in terms of area) with respect to the vertical width and the horizontal width of the reference detection frame. Further, the vertical width and the horizontal width of the fourth non-reference detection frame are set to be three times the vertical width and the horizontal width of the third non-reference detection frame, respectively. Then, the vertical width and the horizontal width of the fifth non-reference detection frame are set to be three times the vertical width and the horizontal width of the fourth non-reference detection frame, respectively.

  When a plurality of types of detection frames of such a size are associated with the used captured image of the (k-1) th frame, the detection process is performed using the third non-reference detection frame. At this time, the vertical and horizontal widths of the used captured image of the (k−1) th frame are set to (1/3) times to generate a size-changed image for the third non-reference detection frame. Further, when performing the detection process using the fourth non-reference detection frame, the size of the third non-reference detection frame is changed instead of reducing the used captured image of the (k−1) frame. The vertical and horizontal widths of the image are set to (1/3) times to generate a resized image for the fourth non-reference detection frame. When performing the detection process using the fifth non-reference detection frame, the size of the fourth non-reference detection frame is changed instead of reducing the use captured image of the (k−1) frame. The vertical and horizontal widths of the images are set to (1/3) times to generate a resized image for the fifth non-reference detection frame. As a result, by performing the reduction process with the same reduction ratio, it is possible to generate the resized images for the third to fifth non-reference detection frames.

  Further, the sixth to eighth non-reference detection frames having different sizes are associated with the used captured image of the k-th frame, and the sizes are smaller in the order of the sixth to eighth non-reference detection frames. Shall. In this case, the vertical width and the horizontal width of the sixth non-reference detection frame are set to, for example, 5 times (25 times in terms of area) with respect to the vertical width and the horizontal width of the reference detection frame. Further, the vertical width and the horizontal width of the seventh non-reference detection frame are set to 5 times the vertical width and the horizontal width of the sixth non-reference detection frame, respectively. Then, the vertical width and the horizontal width of the eighth non-reference detection frame are each set to 5 times the vertical width and the horizontal width of the seventh non-reference detection frame.

  When a plurality of types of detection frames of such a size are associated with the use image of the k-th frame, k is used when performing detection processing using the sixth non-reference detection frame. The vertical and horizontal widths of the used captured image of the frame are set to (1/5) times to generate a resized image for the sixth non-reference detection frame. In addition, when performing detection processing using the seventh non-reference detection frame, the vertical width of the size-changed image for the sixth non-reference detection frame is used instead of reducing the use image of the kth frame. And the horizontal width is set to (1/5) times, and the resized image for the seventh non-reference detection frame is generated. When performing the detection process using the eighth non-reference detection frame, the vertical width of the size-changed image for the seventh non-reference detection frame is used instead of reducing the use image of the k-th frame. And the width is set to (1/5) times, and the resized image for the eighth non-reference detection frame is generated. As a result, by performing the reduction process with the same reduction ratio, it is possible to generate the resized images for the sixth to eighth non-reference detection frames.

  The detection unit 3 performs the above-described detection processing for each of a plurality of use captured images constituting the use captured image group, and a plurality of types of detection frames (in this example, 10 types) associated with the use captured image. Each detection frame). As a result, for each use captured image, at least one detection result region (region that is likely to be a face image) and a detection result frame corresponding to each of a plurality of types of detection frames associated with the use captured image. (Outer frame of an area having a high possibility of being a face image) is obtained, and detection accuracy values corresponding to the respective detection result frames are obtained. The detection accuracy value corresponding to the detection result frame obtained for the used captured image indicates the probability that the image in the detection result frame in the used captured image is a face image.

<Description of operation of detection target image specifying unit>
A plurality of used captured images constituting a used captured image group used when a face image is detected from a processing target image may be considered to be the same image by bringing their imaging timings close to each other. it can.

  In the present embodiment, the imaging frame rate at the imaging unit is 30 fps, and the use captured image group is configured by three captured images continuously captured by the imaging unit. The imaging interval is (1/30) seconds. If the walking speed of a person is 5 km / hour, the distance that the walking person moves in (1/30) seconds is about several centimeters. That is, the human face hardly moves while a plurality of use captured images are captured. Therefore, from the viewpoint of detecting a human face image from the processing target image, it can be seen that the processing target image and the other used captured images are the same image. Therefore, it can be considered that the detection result frame obtained for the used captured image other than the processing target image in the used captured image group is equivalent to the detection result frame obtained for the processing target image. In other words, if the detection result frame obtained for the used captured image other than the processing target image is placed on the processing target image, the image within the detection result frame in the processing target image is an image that is highly likely to be a face image. It can be said that there is. Then, the detection accuracy value for the detection result frame obtained for the use captured image other than the processing target image is the face image of the image within the detection result frame when the detection result frame is placed on the processing target image. It can be said that it shows its uniqueness.

  Therefore, the detection target image specifying unit 7 according to the present embodiment uses all the detection result frames obtained for the used captured images other than the processing target images as detection result frames for the processing target images. That is, the detection target image specifying unit 7 sets all the detection result frames obtained for the used captured image group as detection result frames for the processing target image. Then, the detection target image specifying unit 7 specifies a face image in the processing target image based on each detection result frame for the processing target image and the detection accuracy value for each detection result frame.

  As described above, by using the detection result frame obtained for the used captured image other than the processing target image as the detection result frame for the processing target image, in the detection processing for the processing target image, one of the T types of detection frames. In spite of using only a part, it is possible to obtain a detection result frame equivalent to the detection result frame obtained when the detection process is performed using all of the T types of detection frames. Hereinafter, unless otherwise specified, the detection result frame for the processing target image used in the detection target image specifying unit 7 means all of the detection result frames obtained for the used captured image group.

  FIG. 12 shows a state in which all detection result frames 150 obtained for a plurality of use captured images constituting the use captured image group are arranged to overlap the process target image 20a as detection result frames for the process target image 20a. FIG.

  As shown in FIG. 12, detection results frames 150 of various sizes are obtained by performing detection processing using a plurality of types of detection frames having different sizes. This means that face images of various sizes included in the processing target image 20a are detected. Also, as shown in FIG. 12, when the obtained detection result frame 150 is superimposed on the processing target image 20a, a plurality of detection result frames 150 are concentrated in the vicinity of one face image. That is, a plurality of detection result frames 150 are obtained for one face image included in the processing target image 20a.

  Thus, since a plurality of detection result frames 150 are obtained for one face image included in the processing target image 20a, it is difficult to specify the number of face images included in the processing target image 20a. It is. In addition, if the processing target image 20a with the detection result frame 150 overlapped as shown in FIG. 12 is displayed on the display device, the face image included in the processing target image 20a may be hidden by the plurality of detection result frames 150. And it is difficult to identify the face image.

  Accordingly, a plurality of detection result frames 150 concentrated near the face image are integrated into one detection result frame to generate an integrated detection result frame, and one integrated detection result frame is associated with one face image. Is desirable.

  On the other hand, if the plurality of detection result frames 150 are not properly integrated, the face image does not fit properly in the integrated detection result frame, and as a result, the detection accuracy of the face image may be lowered.

  The detection target image specifying unit 7 according to the present embodiment uses the output value map generated by the map generation unit 4 to specify a face image in the processing target image, and uses the outline frame of the face image as the integrated detection result frame. By doing so, a highly accurate integrated detection result frame, that is, an integrated detection result frame in which the face image is properly contained inside is generated. First, output value map generation processing will be described.

<Output value map generation processing>
The map generation unit 4 generates an output value map indicating the distribution in the processing target image with respect to the detection accuracy value indicating the likelihood (face image likelihood) as the face image based on the detection result of the detection unit 3.

  Specifically, the map generation unit 4 is a map composed of a total of (M × N) values in which M values are arranged in the row direction and N values are arranged in the column direction, similarly to the processing target image. Think about 200. Then, the map generation unit 4 sets one detection result frame for the processing target image as the target detection result frame, and sets a frame 210 having the same size as the target detection result frame to the map 200 at the same position as the target detection result frame. To set. FIG. 13 is a diagram illustrating a state in which a frame 210 is set for the map 200.

  Next, the map generation unit 4 sets “0” for each value outside the frame 210 in the map 200, and for each value within the frame 210, a detection accuracy value corresponding to the target detection result frame (the target detection result frame and The detection accuracy value obtained as a result of detecting the face image with respect to the image within the detection frame is determined. If the size of the target detection result frame is, for example, 16p × 16p, there are 16 values in the row 210, 16 in the column direction and 16 in the column direction, for a total of 256 values. Further, assuming that the size of the target detection result frame is 20p × 20p, for example, there are 20 values in the row 210 and 20 in the column direction, for a total of 400 values. FIG. 14 is a diagram for explaining a method for determining each value in the frame 210.

  The map generation unit 4 sets the value of the center 211 in the frame 210 as the detection accuracy value corresponding to the target detection result frame obtained by the detection unit 3. Then, the map generation unit 4 gradually increases the values of the other values in the frame 210 from the center 211 in the frame 210 to the outside according to a normal distribution curve with the value at the center 211 of the frame 210 as the maximum value. To be smaller. Thereby, each of a plurality of values constituting the map 200 is determined, and the map 200 corresponding to the target detection result frame is completed.

  As described above, the map generation unit 4 generates a plurality of maps 200 respectively corresponding to a plurality of detection result frames for the processing target image. In other words, the map generation unit 4 generates a plurality of maps 200 respectively corresponding to a plurality of detection result frames obtained for a plurality of used captured images constituting the used captured image group. And the map production | generation part 4 synthesize | combines the produced | generated several map 200, and produces | generates an output value map. Specifically, the map generation unit 4 adds the m × n-th values of the plurality of generated maps 200, and uses the obtained addition value as the m × n-th detection accuracy value of the output value map. . In this way, the map generation unit 4 obtains each detection accuracy value constituting the output value map. Thereby, an output value map indicating the distribution of detection accuracy values in the processing target image is completed. By referring to the output value map, it is possible to specify a region having a high likelihood of a face image in the processing target image. That is, the face image in the processing target image can be specified by referring to the output value map.

  FIG. 15 is a diagram showing an output value map for the processing target image 20a shown in FIG. 12 superimposed on the processing target image 20a. In FIG. 15, for easy understanding, the output value map is shown by dividing the magnitude of the detection accuracy value into, for example, five stages from the first stage to the fifth stage. In the output value map shown in FIG. 15, vertical line hatching is shown for the area belonging to the fifth stage where the detection accuracy value is the largest, and sand area is shown for the area belonging to the second largest fourth stage. Hatching is shown. Further, in the output value map in FIG. 15, the area belonging to the third stage where the detection accuracy value is the third largest shows hatching to the right, and the area belonging to the fourth largest second stage. Indicates a hatching that goes up to the left. In the output value map shown in FIG. 15, hatching is not shown for the region belonging to the first stage having the smallest detection accuracy value.

  Referring to FIG. 15, it can be understood that in the output value map, the detection accuracy value in the region corresponding to each face image included in the processing target image 20a is high.

<Maximum point search processing>
As shown in FIG. 15, in the output value map, there is a high possibility that the detection accuracy value in the region corresponding to the face image in the processing target image becomes large. Then, looking at the output value map from a micro viewpoint, in the output value map, the detection accuracy value in the region corresponding to the face image in the processing target image is the same as the center position of the face image. There is a high possibility that the detection accuracy value will be the largest. Therefore, the center position of the face image can be specified by searching for the maximum point of the detection accuracy value in the output value map. Then, by determining that a predetermined area including a pixel at the same position as the identified maximum point (corresponding to the center position of the face image) in the processing target image is a face image, the face image in the processing target image is accurately Can be specified. Therefore, an accurate integrated detection result frame can be obtained by using the outer frame of the predetermined area as the integrated detection result frame.

  Here, a method for searching for the maximum point of the detection accuracy value in the output value map will be described. In the present embodiment, the maximum point search unit 5 searches for the maximum point of the detection accuracy value in the output value map using, for example, the Mean-Shift method. The operation of the maximum point search unit 5 will be described in detail below. Hereinafter, simply speaking, “maximum point” means “maximum point of detection accuracy value in output value map”.

  In the output value map arranged in the two-dimensional coordinates, the local maximum point search unit 5 uses the detection accuracy value as a weighting factor, and coordinates values for a plurality of positions each having a plurality of detection accuracy values included in the processing target region. The maximum point is searched by repeating the process of moving the processing target region so that the center position of the processing target region becomes the weighted average value. In the present embodiment, as many local points as the plurality of detection result frames 150 for the processing target image are obtained.

  FIG. 16 is a diagram for explaining a search method for local maximum points. In FIG. 16, an output value map 300 is arranged at XY coordinates that are two-dimensional coordinates. In the present embodiment, for example, the upper left of the output value map 300 is the origin O of the XY coordinates, the row direction is the X axis direction, and the column direction is the Y axis direction. The shape of the processing target area 400 that is moved when searching for the maximum value is, for example, circular.

  The maximum point search unit 5 sets one detection result frame 150 among the plurality of detection result frames 150 for the processing target image 20a as the target detection result frame 150t.

  Next, the local maximum point search unit 5 determines the movement start position of the processing target area 400 to be moved on the output value map 300. Here, since the image in the detection result frame 150 in the processing target image is highly likely to be a face image, the center of the face image in the processing target image is likely to exist in the detection result frame 150. Therefore, in the output value map 300, there is a high possibility that a local maximum point exists in the region at the same position as the target detection result frame 150t. In particular, in the present embodiment, when the above-described map 200 used for generating the output value map 300 is completed, the value of the center 211 in the frame 210 is used as the detection accuracy value. Is likely to have a local maximum near the same position as the center position in the object detection result frame 150t.

  Therefore, as illustrated in FIG. 16, the local maximum point search unit 5 uses a predetermined position in the target detection result frame 150 t, for example, a position 410 in the output value map 300 that is the same as the center position as the center position of the processing target region 400. The initial position. That is, when starting the search for the maximum point, the local maximum point search unit 5 sets the processing target region 400 so that the center position of the processing target region 400 is the same as the central position in the target detection result frame 150t. 400 is placed in the output value map 300. Thereby, the local maximum point can be searched immediately.

  Note that the size of the processing target area 400 is, for example, such that 50 to 60 detection accuracy values are arranged along the radial direction from the center in the processing target area 400 arranged on the output value map 300. It is a size.

  Next, the maximal point search unit 5 uses the detection accuracy value as a weighting factor in the output value map 300 arranged in the XY coordinates, and a plurality of positions where a plurality of detection accuracy values included in the processing target region 400 exist. A weighted average value (XM, YM) of coordinate values is calculated. The maximum point search unit 5 calculates a weighted average value (XM, YM) using the following equation (1).

  Here, J in Expression (1) indicates the number of a plurality of detection accuracy values existing in the processing target area 400. In addition, i indicates a number assigned to each of a plurality of detection accuracy values in the processing target area 400. And vi means the i-th detection accuracy value, and (Xi, Yi) is the XY coordinate value for the position where the i-th detection accuracy value exists in the output value map 300 arranged in the XY coordinates. Is shown.

  When the maximum point search unit 5 obtains the weighted average value (XM, YM), the processing target region 400 is moved so that the XY coordinates of the center position of the processing target region 400 become the weighted average value (XM, YM). . The arrows in FIG. 16 indicate how the processing target area 400 moves.

  Next, the local maximum point search unit 5 determines whether the moving distance (shift amount) of the processing target area 400 is less than a threshold value. The movement distance of the processing target area 400 is obtained by calculating the distance between the center position of the processing target area 400 before the movement and the center position of the processing target area 400 after the movement. When the local maximum point search unit 5 determines that the movement distance of the processing target area 400 is equal to or greater than the threshold value, the local maximum point searching unit 5 has a plurality of detection accuracy values included in the processing target area 400 after the movement. A weighted average value (XM, YM) of coordinate values is calculated using equation (1). Then, the local maximum point search unit 5 further moves the processing target region 400 so that the XY coordinates of the center position of the processing target region 400 become the newly obtained weighted average value (XM, YM).

  On the other hand, when the local maximum point search unit 5 determines that the movement amount of the processing target region 400 is less than the threshold value, the local maximum point searching unit 5 determines that the movement amount of the processing target region 400 has converged, and moves the processing target region 400. finish. Then, the local maximum point search unit 5 sets the current center position of the processing target region 400 as the local maximum point. From this, the local maximum point in the vicinity of the position of the target detection result frame 150 is obtained.

  As described above, the maximal point search unit 5 obtains a maximal point near the position of the detection result frame 150 for each of the plurality of detection result frames 150 for the processing target image.

  In addition, when calculating the weighted average value (XM, YM), the maximum point search unit 5 may use the output value map 300 from which the detection accuracy value is thinned out. In other words, when calculating the weighted average value (XM, YM), the local maximum point search unit 5 does not have to use all of the plurality of detection accuracy values included in the processing target area 400 in the output value map 300. good.

  FIG. 17 is a diagram illustrating an example of an output value map 300 in which detection accuracy values are thinned out. In the example of FIG. 17, the output value map 300 is a lattice of a plurality of dividing lines 500 arranged at equal intervals in the row direction (X-axis direction) and a plurality of dividing lines 510 arranged at equal intervals in the column direction (Y-axis direction). It is divided into shapes. In the output value map 300, detection accuracy values other than the detection accuracy values existing at the grid intersections (intersections of the dividing lines 500 and 510) are deleted. Thereby, in the output value map 300, (P-1) pieces of detection accuracy values are thinned out every P (P ≧ 2) in the column direction, and every Q pieces (Q ≧ 2) in the row direction ( Q-1) Thinned out. The circles in FIG. 17 indicate the detection accuracy values existing in the output value map 300 from which the detection accuracy values are thinned out.

  In the output value map 300 shown in FIG. 17, similarly to the output value map in FIG. 15, the detection accuracy value is divided into five stages, for example, from the first stage to the fifth stage, and each detection accuracy value is indicated. Has been. In the output value map 300 in FIG. 17, the circle indicating the detection accuracy value belonging to the largest fifth stage is indicated by horizontal hatching, and the circle indicating the detection accuracy value belonging to the second largest fourth stage. The mark shows vertical hatching. Further, in the output value map 300 in FIG. 17, the circle indicating the detection accuracy value belonging to the third largest third stage indicates a right-up hatching, and the detection belonging to the fourth largest second stage. The circle indicating the accuracy value indicates a left-upward hatching. In the output value map 300 shown in FIG. 17, no hatching is shown in the circle indicating the detection accuracy value belonging to the smallest first stage.

  When calculating the weighted average value (XM, YM), the maximum point search unit 5 uses the output value map 300 in which the detection accuracy values are thinned as shown in FIG. In the map 300, the XY coordinate values and the plurality of detection accuracy values for a plurality of positions where the plurality of detection accuracy values included in the processing target area 400 respectively exist are substituted into the above equation (1). This reduces the number of detection accuracy values used when calculating the weighted average value (XM, YM). Furthermore, the number of times of calculating the weighted average value (XM, YM) required until the movement amount of the processing target area 400 converges is also reduced. Therefore, the processing load for searching for the maximum point is reduced.

<Maximum point integration processing>
As can be understood from the above description, the maximum point search unit 5 may require a plurality of maximum points having different positions. In the output value map, a plurality of local maximum points located close to each other is likely to indicate the center of the same face image. On the other hand, there is a high possibility that the plurality of local maximum points located apart from each other indicate the centers of different face images.

  Therefore, when the detection target image determination unit 6 specifies a face image in the processing target image using the maximum point obtained by the maximum point search unit 5, first, the detection target image determination unit 6 obtains the maximum point search unit 5. A plurality of local maximum points located close to each other are integrated into a single local maximum point. Hereinafter, an example of a method for integrating a plurality of local maximum points located close to each other will be described.

  Similar to the case where the detection unit 3 moves the detection frame from the upper left to the lower right of the processing target image in the detection process, the detection target image determination unit 6 looks at the output value map from the upper left to the lower right (raster scanning). When the local maximum obtained by the local maximum searching section 5 appears, the distance between the reference point and the next local maximum appearing is determined using the local maximum as a reference point. Then, if the obtained distance is less than the threshold value, the detection target image determination unit 6 leaves the reference point and deletes the maximum point that appears later. On the other hand, if the obtained distance is equal to or greater than the threshold value, the detection target image determination unit 6 leaves the current reference point and sets the maximum point that appears later as a new reference point.

  The threshold value used in the integration of local maximum points is determined according to how large a face image should be detected. For example, when this image detection apparatus 1 is used in a surveillance camera system and an area relatively close to the camera is monitored, a relatively large face image is detected. A large value is used. In addition, when the image detection apparatus 1 is used in a surveillance camera system and an area relatively far from the camera is monitored, a relatively small face image is detected. A small value is used. In this example, the threshold value is set to a distance of 5 pixels in the processing target image, for example. The threshold value is preferably adjustable (rewritable) by the user.

  If the obtained distance is less than the threshold value, the detection target image determination unit 6 leaves the current reference point, deletes the maximum point that appears later, and then deletes the current reference point and the next maximum point after the deletion. The threshold value is compared with the distance between the local maximum points appearing in. Further, when the obtained distance is equal to or greater than the threshold value and the maximum point that appears later is set as a new reference point, the detection target image determination unit 6 sets the maximum point that appears next to the new reference point and the new reference point. The distance between the points is compared with the threshold value.

  Thereafter, the detection target image determination unit 6 operates in the same manner, and when the distance between the maximum point that appears last and the reference point is less than the threshold value, the maximum point that appears last is deleted, and the maximum point is deleted. This completes the integration process. On the other hand, when the distance between the maximum point that appears last and the reference point is equal to or greater than the threshold value, the detection target image determination unit 6 does not delete the maximum point that appears last, and integrates the maximum points. Exit. The detection target image determination unit 6 specifies a face image in the processing target image using at least one local maximum point remaining after the local maximum point integration process is completed.

  In the above example, if the distance between the reference point and the maximum point that appears later is less than the threshold value, only the reference point is left among the reference point and the maximum point that appears later. Of the local maximum points that appear later, only the local maximum point with the larger detection accuracy value at that position may be left. Accordingly, there is a high possibility that the maximal point remaining after the maximal point integration process ends indicates the center position of the face image.

<Face image determination process>
When the integration process of local maximum points is completed, the detection target image determination unit 6 applies a predetermined region including a pixel at the same position as the local maximum point in the processing target image for each remaining local maximum point after that. It is determined that Then, the detection target image determination unit 6 sets the outline frame of the predetermined area determined to be a face image as a post-integration detection result frame. Hereinafter, the predetermined area determined to be a face image is referred to as a “detected image area”. In addition, the maximum point to be described may be referred to as “target maximum point”.

  Here, in the region corresponding to the face image in the output value map, the detection accuracy value at the same position as the center position of the face image increases, and the detection accuracy value may decrease as the distance from the same position increases. high. In the output value map, there is a high possibility that the detection accuracy value is zero or very small in an area other than the area corresponding to the face image. Therefore, in the output value map, there is a high possibility that the detection accuracy value becomes smaller from the maximum point corresponding to the center position of a certain face image toward the position corresponding to the end of the face image. In other words, in the output value map, there is a high possibility that the detection accuracy value monotonously decreases from the maximum point corresponding to the center position of a certain face image toward the position corresponding to the end of the face image.

  FIG. 18 is a graph showing an example of the distribution of detection accuracy values near the target maximum point 700 in the output value map. FIG. 18 shows a distribution of detection accuracy values in the left-right direction with the target maximum point 700 as the center. In FIG. 18, the vertical axis represents the detection accuracy value, and the horizontal axis represents the position in the left-right direction on the output value map 300. As shown in FIG. 18, the detection accuracy value decreases from the target maximum point 700 in the right direction DR1 (monotonically decreases). Further, the detection accuracy value decreases from the target maximum point 700 in the left direction DR2 (monotonically decreases).

  In the present embodiment, in view of such points, the detection target image determination unit 6 looks at the detection accuracy value from the target maximum point 700 along the direction away from the target maximum point 700 in the output value map. In this case, the position in the processing target image is the same as the position where the detection accuracy value first becomes (1 / Z) times or less (Z> 1) with respect to the detection accuracy value at the target maximum point 700. By defining the edge, the edge of the face image is specified. Thereby, the edge of the face image included in the processing target image can be accurately specified.

  In addition, when a plurality of face images are in contact with each other in a processing target image because a plurality of faces exist in front and back in the imaging area of the imaging unit, the target maximum is along a direction away from the target maximum point 700. When the detection accuracy value is viewed from the point 700, before the position where the detection accuracy value is (1 / Z) times or less than the detection accuracy value at the target maximum point 700 appears, the detection accuracy value The change may not be monotonous. In such a case, there is a high possibility that the position in the processing target image that is the same as the position where the change in the detection accuracy value is not monotonously decreased is the end of the face image.

  Therefore, when the detection target image determination unit 6 looks at the detection accuracy value from the target maximum point 700 along the direction away from the target maximum point 700 in the output value map 300, the detection accuracy value is the target maximum point. If it is determined that the change in the detection accuracy value is no longer monotonically decreasing before a position that is less than (1 / Z) times the detection accuracy value in 700, the position that is determined that the change is no longer monotonically decreasing The position in the same processing target image is the end of the detected image area. Thereby, even when a plurality of face images are in contact with each other, each of the plurality of face images can be specified individually. The operation of the detection target image determination unit 6 will be described in detail below.

  FIG. 19 is a diagram for explaining a method of determining a detection image region 600 (hereinafter, referred to as “target detection image region 600”) including a pixel at the same position as the target maximum point 700 in the processing target image. In FIG. 19, the output value map 300 is shown enlarged. Also, in FIG. 19, the outer shape frame 600 a of the target detection image region 600 is shown superimposed on the output value map 300.

  In the present embodiment, the shape of the detected image area is set to, for example, a quadrangle. The detection target image determination unit 6 determines the detection image region by determining the right end, left end, upper end, and lower end of the square detection image region.

  First, the operation of the detection target image determination unit 6 when the detection target image determination unit 6 determines the right end 610 of the target detection image region 600 will be described. FIG. 20 is a flowchart showing the operation.

  As illustrated in FIG. 19, the detection target image determination unit 6 looks at the detection accuracy value 800 (indicated by a circle) from the target maximum point 700 along the right direction DR1 in the output value map 300. Then, the two detection accuracy values 800 are compared before and after changing the pair, and the right end 610 of the target detection image region 600 is determined based on the comparison result. At this time, the detection target image determination unit 6 may view the detection accuracy values 800 one by one, every other one, or every other plurality. In this example, it is assumed that the detection target image determination unit 6 looks at the detection accuracy values 800 one by one.

  Specifically, as shown in FIG. 20, the detection target image determination unit 6 first sets the detection accuracy value 800 (the detection accuracy value 800a shown in FIG. 19) at the target maximum point 700 to the first in step s11. And the detection accuracy value 800 on the right side thereof is set as the second accuracy value v2. In step s12, the detection target image determination unit 6 obtains (v1-v2) and compares the first accuracy value v1 and the second accuracy value v2.

  Next, in step s13, the detection target image determination unit 6 determines whether the comparison result between v1 and v2 is (v1-v2) <0. If the detection target image determination unit 6 determines that (v1−v2) <0 is not satisfied, the detection target image determination unit 6 determines that the detection accuracy value 800 is monotonically decreasing and executes step s14. In step s14, the detection target image determination unit 6 determines whether the second accuracy value v2 is equal to or less than (1 / Z) times the detection accuracy value 800 at the target maximum point 700. For Z, for example, 3 ≦ Z ≦ 5 is set. When the detection target image determination unit 6 determines that the second accuracy value v2 is equal to or less than (1 / Z) times the detection accuracy value 800a at the target maximum point 700, in step s15, the output value map 300 The position in the processing target image that is the same as the position 710 (see FIGS. 18 and 19) where the accuracy value v2 of 2 exists is the right end 610 of the target detection image area 600. This position 710 indicates that when the detection accuracy value 800 is viewed along the right direction DR1 from the target maximum point 700 in the output value map 300, the detection accuracy value 800 becomes the detection accuracy value 800a at the target maximum point 700. First, it is a position that becomes (1 / Z) times or less.

  On the other hand, in step s14, if the detection target image determination unit 6 determines that the second accuracy value v2 is not less than (1 / Z) times the detection accuracy value 800a at the target maximum point 700, in step s11, The current second accuracy value v2 is set as a new first accuracy value v1, and the detection accuracy value 800 on the right side thereof is set as a new second accuracy value v2. Thereafter, the detection target image determination unit 6 operates in the same manner.

  In step s13, when the detection target image determination unit 6 determines that the comparison result between v1 and v2 is (v1-v2) <0, executes step s16. In step s16, the detection target image determination unit 6 determines whether the comparison result of (v1-v2) <0 has been generated a predetermined number of times C (C ≧ 2) including the current comparison result. That is, when the detection target image determination unit 6 compares the two detection accuracy values 800 before and after changing the pair along the right direction DR1 from the target maximum point 700 in the output value map 300, the previous detection is performed. It is determined whether a comparison result that the accuracy value 800 is smaller than the subsequent detection accuracy value 800 has occurred a predetermined number of times C continuously. The predetermined number of times C is set to C = 2, for example.

  In step s16, when the detection target image determination unit 6 determines that the comparison result of (v1-v2) <0 has continuously occurred a predetermined number of times C, the detection accuracy value is the detection accuracy value at the target maximum point 700. Before a position that is (1 / Z) times or less appears, it is determined that the change in the detection accuracy value is no longer monotonically decreasing, and step s15 is executed, so that the current second accuracy value v2 exists. The position in the processing target image that is the same as the position 710 to be processed is set as the right end 610 of the target detection image area 600. This position 710 indicates that when the detection target image determination unit 6 looks at the detection accuracy value 800 along the right direction DR1 from the target maximum point 700 in the output value map 300, the detection accuracy value 800 is the target maximum. Before a position that is (1 / Z) times or less than the detection accuracy value 800a at the point 700 appears, the detection accuracy value is determined to be no longer monotonically decreasing.

  On the other hand, in step s16, if the detection target image determination unit 6 does not determine that the comparison result of (v1-v2) <0 has continuously occurred a predetermined number of times C, the current second image is determined in step s11. The accuracy value v2 is set as a new first accuracy value v1, and the detection accuracy value 800 on the right side thereof is set as a new second accuracy value v2. Thereafter, the detection target image determination unit 6 operates in the same manner.

  In this way, the detection target image determination unit 6 determines the right end 610 of the target detection image region 600.

  Similarly, when the detection target image determination unit 6 determines the left end 620 of the target detection image region 600, as shown in FIG. 19, in the output value map 300, the left direction DR2 from the target maximum point 700 is displayed. , The two detection accuracy values 800 are compared before and after changing the pair, and the left end 620 of the target detection image region 600 is determined based on the comparison result. When the right end 610 and the left end 620 of the target detection image region 600 are determined, the left-right direction (row direction) width W1 (see FIGS. 18 and 19) of the target detection image region 600 is determined.

  Further, when the detection target image determination unit 6 determines the upper end 630 of the target detection image region 600, as shown in FIG. 19, in the output value map 300, the detection target image determination unit 6 extends from the target maximum point 700 along the upward direction DR3. Then, the two detection accuracy values 800 are compared while changing the pair, and the upper end 630 of the target detection image region 600 is determined based on the comparison result. When the detection target image determination unit 6 determines the lower end 640 of the target detection image region 600, as shown in FIG. 19, in the output value map 300, the target maximum point 700 is shifted downward DR4. The two detection accuracy values 800 are compared while changing the pair, and the lower end 640 of the target detection image region 600 is determined based on the comparison result. When the upper end 630 and the lower end 640 of the target detection image area 600 are determined, the vertical width (column direction) width W2 (see FIG. 19) of the target detection image area 600 is determined.

  In this way, the detection target image determination unit 6 determines the position and size of the detection image region by determining the right end, left end, upper end, and lower end of the quadrangle detection image region. Then, the detection target image determination unit 6 sets the determined outer frame of the detected image region as a post-integration detection result frame. An image in the post-integration detection result frame in the processing target image area becomes a detected image area determined to be a face image.

  The detection target image determination unit 6 determines a detection image area (a face image in the processing target image) corresponding to the local maximum point for each local maximum point remaining after the local maximum point integration process is completed, and the detection image. The outer frame of the region is used as a detection result frame after integration. Thereby, for each face image included in the processing target image, one post-integration detection result frame is obtained for one face image.

  In addition, when the size of the obtained detection image area is too small, the detection target image determination unit 6 may delete the detection image area, assuming that the detection image area is not a face image. In other words, if the size of the obtained detection result frame after integration is too small, the detection target image determination unit 6 determines that the image in the detection result frame after integration is not a face image, and the detection result after integration. The frame may be deleted.

  FIG. 21 shows the detection image region 600 and the post-integration detection result frame 900 (the outer shape frame 600a of the detection image region 600) obtained by the detection target image determination unit 6 with respect to the processing target image 20a shown in FIGS. FIG. In FIG. 21, the detection image region 600 and the post-integration detection result frame 900 are shown superimposed on the processing target image 20a.

  As shown in FIG. 21, approximately one detected image area 600 is obtained for each face image included in the processing target image 20a. That is, approximately one post-integration detection result frame 900 is obtained for each face image included in the processing target image 20a. This is because a plurality of detection result frames 150 (see FIG. 12) obtained for one face image are integrated, and one post-integration detection result frame 900 is obtained for the one face image. Means. In addition, the face images are appropriately contained in each post-integration detection result frame 900. Therefore, it can be said that the face image is appropriately detected in the image detection apparatus 1 according to the present embodiment.

  As described above, in the present embodiment, the processing target image is the same as the maximum point of the detection accuracy value in the output value map indicating the distribution in the processing target image with respect to the detection accuracy value indicating the probability as the detection target image. It is determined that the predetermined area including the pixel at the position is the detection target image. Since the maximum point of the detection accuracy value in the output value map is considered to correspond to the center position of the detection target image in the processing target image, the predetermined region including the pixel at the same position as the maximum point in the processing target image Is determined to be a detection target image, the detection target image can be accurately detected from the processing target image. That is, the detection accuracy for the detection target image can be improved.

  In the above example, step s16 is executed in order to suppress erroneous determination that the monotonous decrease is no longer caused by the influence of noise, but step s16 need not be executed. In this case, if it is determined in step s13 that (v1-v2) <0, step s15 is executed. That is, when the comparison result of (v1-v2) <0 is obtained even once, before the position where the detection accuracy value becomes (1 / Z) times or less than the detection accuracy value at the target maximum point 700 appears. At the same time, it is determined that the change in the detection accuracy value is not monotonously decreased, and step s15 is executed.

  In addition, when the processing target image is displayed on the display device, the image detection device 1 displays the post-integration detection result frame 900 (the outer shape frame 600a of the detection image region 600) for the processing target image, as shown in FIG. ) May be displayed in an overlapping manner.

  In addition, the image detection apparatus 1 compares the face image registered in advance with the detection image region 600 (image in the post-integration detection result frame 900) determined to be a face image in the processing target image. It may be determined whether or not they match. If the face image registered in advance and the detected image area 600 in the processing target image do not match, the image detecting apparatus 1 performs mosaic processing on the detected image area 600 and then The processing target image may be displayed on the display device. As a result, when the image detection apparatus 1 according to the present embodiment is used in a surveillance camera system, even when a face image of a neighbor's person is photographed by the surveillance camera, the face image cannot be recognized. can do. That is, a privacy mask can be realized.

  As described above, in the present embodiment, the T types of detection frames are distributed and associated with the K used captured images including the processing target images that constitute the used captured image group. Less than T types of detection frames are associated with the processing target image. Therefore, detection processing is performed on the processing target image using fewer types of detection frames than T types. Therefore, when the face image is detected from the processing target image, only the processing target image is used, and the processing target image is compared with the case where the detection processing is performed using the T types of detection frames for the processing target image. The amount of processing can be reduced.

  Various effects can be obtained by reducing the processing amount of the processing target image. For example, in the image detection apparatus 1, when each captured image (each frame image) captured by the imaging unit is a processing target image, that is, a facial image of each captured image captured by the imaging unit. When detection is performed, the processing amount for each captured image can be reduced. Therefore, it is possible to reduce the processing amount of face image detection performed in the image detection apparatus 1.

  In the above example, since only (T / 3) types of detection frames among the T types of detection frames are used as the processing target images, when each captured image is a processing target image, Only (T / 3) types of detection frames are used for the image. Therefore, compared with the case where all of the T types of detection frames are used for each captured image, the processing amount in the detection unit 3 is (1/3) times, and the processing amount can be reduced.

  Further, in the image detection device 1, when the captured image captured by the imaging unit is a processing target image for each U (U> K), that is, whenever K or more captured images are obtained by the imaging unit. In addition, when a face image is detected from a captured image, unlike the present embodiment, if the detection processing is performed using all of the T types of detection frames for the processing target image, the image capturing unit obtains the detection result. Of the captured images to be processed, only the processing target image has a very large processing amount. On the other hand, as in the present embodiment, the detection processing using the T types of detection frames is performed in a distributed manner on the K used captured images, so that the processing amount is increased only for the processing target image. Can be suppressed. Therefore, a difference in processing amount between captured images can be reduced.

  In the above example, K = 3. However, K = 2 or K ≧ 4 may be used. In the above example, among the K used captured images constituting the used captured image group, the last captured captured image is used as the processing target image. However, other used captured images may be used as the processing target image. good. For example, the used captured image captured first among the K used captured images constituting the used captured image group may be set as the processing target image.

<Various modifications>
Hereinafter, various examples of the present embodiment will be described by taking as an example a case in which the use captured image group includes the (k-2) th frame, the (k-1) th frame, and the kth frame. A modification will be described.

<First Modification>
Although the plurality of used captured images constituting the used captured image group are captured at timings close to each other, there are some differences between the processing target image and the other used captured images. Therefore, considering only the point of improving the detection accuracy of the face image, it is desirable that the detection process is performed on the processing target image using as many types of detection frames as possible.

  Therefore, in the present modification, the detection unit 3 uses a detection frame associated with the use-captured image for the use-captured image other than the processing target image in the plurality of use-captured images constituting the use-captured image group. As a result of performing detection processing using the detection frame, if there is a region that is very likely to be a face image of the same size as the detection frame in the used captured image, the detection frame is also used for the processing target image. Detection process.

  For example, the detection unit 3 includes, in at least one detection result frame obtained for the use captured image of the (k-2) th frame, whether a detection result frame corresponding to the detection accuracy value corresponding to the threshold value is greater than or equal to a threshold value. Determine. The threshold value (hereinafter referred to as “second threshold value”) is a threshold value (hereinafter referred to as “first threshold”) used when the detection result frame is specified by the determination unit 33. The value is larger than the value).

  When the detection unit 3 identifies a detection result frame whose detection accuracy value corresponding to the use captured image of the (k-2) frame is equal to or greater than the second threshold value, the use imaging of the (k-2) frame is used. It is determined that the image within the detection result frame in the image is a region that is very likely to be a face image having the same size as the detection result frame in the (k-2) -th frame used captured image. That is, when the detection unit 3 specifies a detection result frame having a detection accuracy value corresponding to the second threshold value or more for the use captured image of the (k-2) th frame, the (k-2) th frame In the used captured image, it is determined that there is a region that is very likely to be a face image having the same size as the detection frame corresponding to the detection result frame (a detection frame having the same size as the detection result frame). Then, the detection unit 3 uses the detection frame corresponding to the detection result frame whose detection accuracy value is greater than or equal to the second threshold for the use captured image of the (k-2) frame to the processing target image. Perform detection processing.

  As described above, the detection unit 3 performs the detection process on the use captured image of the (k−2) th frame using the detection frame associated with the use captured image, and as a result, the detection accuracy value is the second threshold value. When the above detection result frame is obtained, it is determined that there is a region in the use captured image of the (k-2) th frame that is very likely to be a face image having the same size as the detection frame. The detection process is also performed on the image using the detection frame. In the detection target image specifying unit 7, the face image in the processing target image is specified using the detection result frame and the detection accuracy value obtained by the detection processing.

  Similarly, the detection unit 3 performs the detection process on the used captured image of the (k−1) th frame using the detection frame associated therewith, and the detection accuracy value is the second threshold value. When the above detection result frame is obtained, it is determined that there is an area in the use captured image of the (k−1) th frame that is very likely to be a face image of the same size as the detection frame, and the processing target The detection process is also performed on the image using the detection frame. In the detection target image specifying unit 7, the face image in the processing target image is specified using the detection result frame and the detection accuracy value obtained by the detection processing.

  As described above, in this modified example, as a result of the detection processing performed on the use captured image other than the processing target image using the detection frame associated with the use captured image, If there is a region that is very likely to be a face image having the same size as the detection frame, the detection frame is also used for the processing target image and the detection process is performed. Therefore, when the face image is specified in the processing target image, the face image is specified with high accuracy by using the result of the detection process. Therefore, the detection accuracy of the face image is improved.

<Second Modification>
As shown in FIG. 12 described above, all the detection result frames 150 obtained for a plurality of use captured images constituting the use captured image group are superimposed on the process target image 20a as detection result frames for the process target image 20a. Are arranged, a plurality of detection result frames 150 are concentrated in the vicinity of one face image. That is, detection result frames 150 obtained for a plurality of used captured images constituting the used captured image group are concentrated near one face image. Therefore, in the processing target image 20a on which the detection result frames 150 are overlapped, even if the detection result frame 150 is present, the use result in which the detection result frame 150 overlapping the region is obtained in a plurality of use captured images. When the number of captured images is small, the possibility that the area is a face image is low. Therefore, the detection result area corresponding to the detection result frame 150 that overlaps the area can be considered to be an area that is erroneously determined to be a face image. Therefore, it is preferable that the detection result frame 150 overlapping the area is not used when the detection target image is specified in the processing target image.

  Therefore, in the present modification, the detection target image specifying unit 7 divides the processing target image 20a on which the detection result frames 150 are superimposed as shown in FIG. 12 into a plurality of blocks. Then, the detection target image specifying unit 7 includes, for a block that overlaps the detection result frame 150 in a plurality of blocks, among the plurality of used captured images, the number of used captured images from which the detection result frame 150 that overlaps the block is obtained. Is equal to or less than the threshold value, the face image is specified in the processing target image 20a without using the detection result frame 150 overlapping the block. In the present modification, the detection target image specifying unit 7 has, for a block overlapping the detection result frame 150, among the plurality of used captured images, the number of used captured images from which the detection result frame 150 overlapping the block is obtained. For example, when it is 1 or less, that is, 1, the face image is specified in the processing target image 20a without using the detection result frame 150 that overlaps the block. Below, this modification is demonstrated concretely. In the following description, the detection result frame 150 obtained for the used captured image of the kth frame is referred to as “detection result frame 150a”, and the detection result frame 150 obtained for the used captured image of the (k−1) th frame is represented by “ The detection result frame 150b ”is obtained, and the detection result frame 150 obtained for the used captured image of the (k-2) th frame is assumed to be a“ detection result frame 150c ”.

  FIG. 22 is a diagram illustrating an example of a state in which the processing target image 20 a on which the detection result frames 150 are superimposed is divided into a plurality of blocks 950. In the example of FIG. 22, the processing target image 20a is divided into nine blocks 950 in the row direction and is divided into seven blocks 950 in the column direction. Note that the method of dividing the processing target image 20a is not limited to this.

  In a block 950a among the plurality of blocks 950, the detection result frame 150a, the detection result frame 150b, and the detection result frame 150c overlap. Therefore, for the block 950a, of the three used captured images constituting the used captured image group, the number of used captured images in which the detection result frame 150 overlapping with the block 950a is obtained is “3”, and the threshold is set. The value is larger than “1”. Therefore, the detection target image specifying unit 7 uses all of the detection result frames 150 overlapping the block 950a when specifying a face image in the processing target image 20a.

  In addition, in the block 950b among the plurality of blocks 950, the detection result frame 150a and the detection result frame 150b overlap. Therefore, for the block 950b, out of the three used captured images constituting the used captured image group, the number of used captured images in which the detection result frame 150 overlapping the block 950b is obtained is “2”, and the threshold is set. The value is larger than “1”. Therefore, the detection target image specifying unit 7 uses all of the detection result frames 150 overlapping the block 950b when specifying the face image in the processing target image 20a.

  On the other hand, in the block 950c of the plurality of blocks 950, only the detection result frame 150a overlaps. Therefore, with respect to the block 950c, of the three used captured images constituting the used captured image group, the number of used captured images in which the detection result frame 150 overlapping the block 950b is obtained is “1”, and the threshold is set. The value is “1” or less. Therefore, the detection target image specifying unit 7 determines that the detection result frame 150 overlapping the block 950c is the detection result frame 150 that is erroneously acquired due to the influence of noise, and the face in the processing target image 20a. Do not use when identifying images.

  In the example of FIG. 22, one detection result frame 150a, one detection result frame 150b, and one detection result frame 150c overlap each other in the block 950a, but actually, a plurality of detection result frames 150a. The plurality of detection result frames 150b and the plurality of detection result frames 150c are likely to overlap. Similarly, in the block 950b, one detection result frame 150a and one detection result frame 150b overlap, but actually, a plurality of detection result frames 150a and a plurality of detection result frames 150b overlap. Probability is high. Similarly, one detection result frame 150a overlaps the block 950c, but in reality, there is a high possibility that a plurality of detection result frames 150a overlap. Even when a plurality of detection result frames 150a overlap with the block 950c, the use captured image in which the detection result frame 150 overlapping with the block 950b is obtained among the three use captured images constituting the use captured image group. Since the number of detection results is 150, all of the plurality of detection result frames 150a overlapping the block 950a are not used when specifying a face image in the processing target image 20a.

  As described above, in this modification, the detection target image specifying unit 7 obtains the detection result frame 150 that overlaps the block among the plurality of used captured images for the block that overlaps the detection result frame 150 in the plurality of blocks. When the number of used captured images is equal to or less than the threshold value, the face image is specified in the processing target image 20a without using the detection result frame 150 that overlaps the block. Therefore, a result of erroneously determining that an image having a feature amount that closely resembles the feature amount of the face image accidentally exists in the used captured image due to the influence of noise, and that the image is likely to be a face image. Even if the detection result frame 150 obtained by the above is present, the detection target image specifying unit 7 can specify the face image in the processing target image by removing the detection result frame 150. Therefore, the detection accuracy of the face image is improved.

  Note that the detection target image specifying unit 7 determines that the number of used captured images in which the detection result frame 150 overlapping the block is obtained is equal to or less than the threshold among the plurality of used captured images for the block overlapping the detection result frame 150. Even when the detection result area having the detection result frame that overlaps the block as the outer frame includes a detection result area that is very likely to be a face image, the detection that overlaps the block is detected. The result frame 150 may be used to specify a face image in the processing target image.

  For example, consider a case where a plurality of detection result frames 150a overlap the block 950c in the example of FIG. In such a case, the detection target image specifying unit 7 has a detection accuracy value corresponding to a third threshold value (> first threshold value) or more in a plurality of detection result frames 150a overlapping the block 950c. It is determined whether or not there is a detection result frame 150a. Then, when the detection target image specifying unit 7 has a plurality of detection result frames 150a overlapping the block 950c and a detection result frame 150a corresponding to the detection accuracy value equal to or greater than the third threshold value, the detection target image specifying unit 7 The detection result area having the detection result frame 150a overlapping the block 950c as the outer frame includes a detection result area very likely to be a face image, and a plurality of detection results overlapping the block 950a A face image is specified in the processing target image using the frame 150a. Thereby, when the face image is specified in the processing target image, it is possible to prevent the correct detection result frame 150 from being erroneously used.

  On the other hand, the detection target image specifying unit 7 determines that, in the plurality of detection result frames 150a overlapping the block 950c, there is no detection result frame 150a having a corresponding detection accuracy value equal to or greater than the third threshold value. Determines that the detection result region having the detection result frame 150a that overlaps the block 950c as an outer frame does not include a detection result region that is very likely to be a face image, and thus includes a plurality of overlapped blocks 950a. A face image is specified in the processing target image without using the detection result frame 150a.

<Third Modification>
Although the plurality of used captured images constituting the used captured image group are captured at timings close to each other, there are some differences between the processing target image and the other used captured images. Therefore, it can be said that the detection accuracy value obtained for the used captured image other than the processing target image has low accuracy when viewed as the detection accuracy value for the processing target image. And about the detection accuracy value obtained about the use captured image other than the processing target image, when the detection timing value of the processing target image is viewed as the imaging timing of the use captured image becomes far from the imaging timing of the processing target image Can be said to be less accurate.

  Therefore, in the present modification, the detection target image specifying unit 7 weights the detection accuracy values corresponding to the detection result frames (detection result regions) obtained for the plurality of used captured images constituting the used captured image group. And a face image is specified in the processing target image based on the detection accuracy value. Specifically, the map generation unit 4 of the detection target image specifying unit 7 weights the detection accuracy values corresponding to the detection result frames obtained for the plurality of used captured images, and then performs the detection accuracy value. An output value map is generated based on Then, when weighting the detection accuracy value corresponding to the detection result frame obtained for the used captured image, the map generation unit 4 sets the capturing timing of the used captured image to be higher than the capturing timing of the processing target image. The farther away, the smaller the weight for the detection accuracy value.

  For example, when generating the map 200 (see FIG. 13) for the detection result frame obtained for the used captured image of the (k−2) th frame, the map generation unit 4 according to the present modification includes the map 200. The set value of the center 211 in the frame 210 (see FIG. 14) is a value obtained by multiplying the detection accuracy value corresponding to the detection result frame by “0.8”. Then, the map generation unit 4 gradually increases the values of the other values in the frame 210 from the center 211 in the frame 210 to the outside according to a normal distribution curve with the value at the center 211 of the frame 210 as the maximum value. To be smaller.

  Further, when the map generation unit 4 generates the map 200 for the detection result frame obtained for the used captured image of the (k−1) th frame, the map generation unit 4 displays the center 211 in the frame 210 set in the map 200. The value is a value obtained by multiplying the detection accuracy value corresponding to the detection result frame by “0.9”. Then, the map generation unit 4 gradually increases the values of the other values in the frame 210 from the center 211 in the frame 210 to the outside according to a normal distribution curve with the value at the center 211 of the frame 210 as the maximum value. To be smaller.

  When the map generation unit 4 generates the map 200 for the detection result frame obtained for the used captured image of the k-th frame that is the processing target image, the center 211 in the frame 210 set in the map 200 is displayed. Is a value obtained by multiplying the detection accuracy value corresponding to the detection result frame by “1.0”. Then, the map generation unit 4 gradually increases the values of the other values in the frame 210 from the center 211 in the frame 210 to the outside according to a normal distribution curve with the value at the center 211 of the frame 210 as the maximum value. To be smaller.

  As described above, when weighting is performed on the detection accuracy value for the detection result region obtained for the used captured image, the farther the imaging timing of the used captured image is from the imaging timing of the processing target image, By reducing the weighting on the detection accuracy value, a highly accurate output value map can be generated. Therefore, it is possible to accurately specify the face image in the processing target image. As a result, the detection accuracy of the face image is improved.

  The first to third modified examples described above can be used in combination with at least two modified examples. For example, the first modified example and the second modified example can be used in combination, and the first to third modified examples can be used in combination.

  Moreover, although the image detection apparatus 1 was demonstrated in detail above, the above-mentioned description is an illustration in all the aspects, Comprising: This invention is not limited to it. The various examples described above can be applied in combination as long as they do not contradict each other. And it is understood that the countless modification which is not illustrated can be assumed without deviating from the scope of the present invention.

DESCRIPTION OF SYMBOLS 1 Image detection apparatus 3 Detection part 7 Detection target image specific part 12 Control program

Claims (10)

An image detection device for detecting a detection target image from a processing target image,
A detection unit that performs a detection process for detecting, as a detection result region, a region that is highly likely to be the detection target image having the same size as the detection frame, using the detection frame;
A detection target image specifying unit that specifies the detection target image in the processing target image,
The detection unit includes at least one type of detection frame included in a plurality of types of detection frames having different sizes with respect to each of a plurality of captured images including the processing target image captured at different timings. In correspondence with each other, the plurality of types of detection frames are distributed and associated with the plurality of captured images, and for each of the plurality of captured images, at least one type of detection frame corresponding to the captured image. The detection process is performed using
The said detection target image specific | specification part is an image detection apparatus which specifies the said detection target image in the said process target image based on the said detection result area | region detected about the said several picked-up image in the said detection part. The image detection apparatus according to claim 1,
The plurality of types of detection frames include a reference detection frame of a reference size and a non-reference detection frame of a size different from the reference size,
The detector is
When performing the detection process for the captured image using the non-reference detection frame,
Resizing the non-reference detection frame so that the size matches the reference size, and changing the size of the captured image according to the size change of the non-reference detection frame,
While moving the size change detection frame that is the non-reference detection frame after the size change to the size change image that is the captured image after the size change, the image within the size change detection frame in the size change image An image detection device that determines whether or not the image is likely to be the detection target image. The image detection apparatus according to any one of claims 1 and 2,
In the captured image, the detection unit performs the detection process on the captured images other than the processing target image in the plurality of captured images using the detection frames included in the multiple types of detection frames. An image detection apparatus that performs the detection process on the processing target image using the detection frame when there is a region that is very likely to be the detection target image having the same size as the detection frame. The image detection apparatus according to any one of claims 1 to 3,
The detection target image specifying unit specifies the detection target image in the processing target image based on a detection result frame that is an outer frame of the detection result region detected for the plurality of captured images in the detection unit. ,
The detection target image specifying unit divides the processing target image on which the detection result frame is overlaid into a plurality of blocks,
The detection target image specifying unit is
About the block that overlaps the detection result frame in the plurality of blocks,
When the number of the captured images obtained from the plurality of captured images that are overlapped with the block is less than or equal to the threshold value, without using the detection result frames that overlap the block, An image detection device that identifies the detection target image in the processing target image. The image detection apparatus according to claim 4,
The detection target image specifying unit is
About the block that overlaps the detection result frame in the plurality of blocks,
Among the plurality of captured images, even if the number of captured images from which the detection result frame overlapping the block is obtained is equal to or less than a threshold value, the detection result frame overlapping the block is defined as an outer frame. When the detection result area that is very likely to be the detection target image is included in the detection result area, the detection result frame that overlaps the block is used to detect the detection in the processing target image. An image detection device that identifies a target image. An image detection device according to any one of claims 1 to 5,
The detection target image specifying unit is based on a detection accuracy value indicating the probability that the detection result region is the detection target image for the detection result region detected for the plurality of captured images in the detection unit. An image detection device that identifies the detection target image in the processing target image. The image detection device according to claim 6,
The detection target image specifying unit weights the detection accuracy value for the detection result region detected for the plurality of captured images in the detection unit, and then based on the detection accuracy value, Specifying the detection target image in the processing target image;
When the detection target image specifying unit weights the detection accuracy value for the detection result area detected for the captured image, the capturing timing of the captured image is higher than the capturing timing of the processing target image. An image detection apparatus that reduces the weighting of the detection accuracy value as the distance increases. An image detection apparatus according to any one of claims 1 to 7,
The image detection apparatus, wherein the detection target image is a human face image. A control program for controlling an image detection device that detects a detection target image from a processing target image,
In the image detection device,
(A) performing a detection process for detecting, as a detection result region, a region that is highly likely to be the detection target image having the same size as the detection frame, using the detection frame;
(B) executing the step of identifying the detection target image in the processing target image;
In the step (a), at least one type of detection included in a plurality of types of detection frames having different sizes with respect to each of a plurality of captured images including the processing target image captured at different timings. The plurality of types of detection frames are distributed and associated with the plurality of captured images so that the frames correspond, and for each of the plurality of captured images, at least one type of detection frame corresponding to the captured image The detection process is performed using
Control for operating in the step (b) to specify the detection target image in the processing target image based on the detection result area detected for the plurality of captured images in the step (a). program. An image detection method for detecting a detection target image from a processing target image,
(A) performing a detection process for detecting, as a detection result region, a region that is highly likely to be the detection target image having the same size as the detection frame, using the detection frame;
(B) specifying the detection target image in the processing target image,
In the step (a), at least one type of detection included in a plurality of types of detection frames having different sizes with respect to each of a plurality of captured images including the processing target image captured at different timings. The plurality of types of detection frames are distributed and associated with the plurality of captured images so that the frames correspond, and for each of the plurality of captured images, at least one type of detection frame corresponding to the captured image The detection process is performed using
An image detection method in which, in the step (b), the detection target image is specified in the processing target image based on the detection result area detected for the plurality of captured images in the step (a).