Object detection device

Abstract

To provide a technology for improving accuracy of object detection based on image information.SOLUTION: Imaging means 20 outputs image information representing an original image. Image division means 11b divides the original image represented by the image information into a plurality of subdivided images. Image processing means 11a detects an object included in each of the subdivided images on the basis of subdivided image information representing the subdivided images. Detection result synthesizing means 11c synthesizes results of object detection on each of the subdivided images, to be output as a detection result.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to an object detection device that detects an object based on image information. [Background technology]

[0002] Technology to detect objects using deep learning (machine learning method using multi-layer neural networks) is being researched. For example, many end-to-end methods have been proposed, such as SSD (Single Shot MultiBox Detector) and YOLO (You Only Look Once), which simultaneously detect the location and category of an object based on image information indicating an image captured by an imaging means. These methods are based on multi-task learning, which simultaneously performs learning using a multi-layer neural network for object location detection and learning using a multi-layer neural network for object category discrimination. An object detection technique using SSD is disclosed in, for example, Non-Patent Document 1, and an object detection technique using YOLO is disclosed in, for example, Non-Patent Document 2. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] “SSD: Single Shot MulitiBox Detector”, Wei Liu, Dragomir Anguelov. Domitru Erhan, Christian Szegedy, Scott reed, Cheng-Yang Fu and Alexsander C. berg (2015), https: / / arxiv.org / pdf / 1512.02325.pdf [Non-Patent Document 2] “You Only Look Once Unified, Real-Time Object Detection”, Joseph Redmon, Santosh Divvala, Ross Girshick and Ali Farhadi (2016), https: / / pjreddie,com / media / files / papers / yolo.pdf Summary of the Invention [Problem to be solved by the invention]

[0004] In recent years, the performance of imaging means has improved, and the number of pixels (resolution) of image information showing captured images tends to increase. For example, currently, many surveillance cameras are compatible with 2K (2 million pixels) or less, but in the future, it is expected that cameras compatible with 4K (8 million pixels) or 8K (33 million pixels) will become more widespread. If imaging means with a large number of pixels (high resolution) can be used, high-definition image information can be obtained, and the accuracy of object detection will improve. On the other hand, current object detection devices are configured to process image information with a resolution of less than 2K (e.g., several hundred by several hundred pixels). For this reason, when current object detection devices process image information with a resolution of 4K, 8K, or higher, there is a risk that detection errors will increase and object detection accuracy will decrease. Configuring current object detection devices to be able to process high-resolution image information requires a great deal of effort and cost. Therefore, the present inventors have conducted various studies on techniques that can improve the accuracy of object detection using current object detection devices, and have found that it is possible to easily and inexpensively improve the accuracy of object detection by dividing image information into a plurality of pieces of divided image information and processing the divided image information. An object of the present invention is to provide a technique that can easily and inexpensively improve the accuracy of object detection by an object detection device. [Means for solving the problem]

[0005] The object detection device of the present invention includes an imaging means and an object detection means. The imaging means outputs image information indicating the captured original image. As the imaging means, a known imaging means constituted by a CCD element or the like can be used. The object detection means includes an image processing means and an image division means. And, It has. The image dividing means divides the original image indicated by the image information output from the imaging means into a plurality of divided images. Any suitable method can be used for dividing the original image into a plurality of divided images (the number of divided images, the size of the divided images, the number of divisions, etc.), but a method of dividing at equal intervals in the vertical and horizontal directions is preferably used, and a method of dividing at equal intervals in the vertical and horizontal directions by the same number of divisions is more preferably used. The image processing means includes: Based on image information indicating the original image, an object detection process is performed on the original image to detect an object contained in the original image, and if the object detection result for the original image contains an object, the object detection result for the original image is output.If the object detection result for the original image does not contain an object, an object detection process is performed on each divided image based on divided image information indicating each divided image to detect an object contained in each divided image, and the object detection result for each divided image is output. As image processing means, well-known image processing means such as SSD and YOLO, which use deep learning to simultaneously detect the position and category of an object based on image information, are used. Used. According to the present invention, the accuracy of object detection by an object detection device can be improved easily and inexpensively. In another embodiment of the present invention, the image division means divides the original image into at least a first number of first divided images and a second number of second divided images. The first number and the second number are set so that the boundary portion of the first divided image and the boundary portion of the second divided image do not overlap in parallel. In this embodiment, the boundary portion of the first divided image and the boundary portion of the second divided image are allowed to intersect. This allows, for example, an object that exists across the boundary portion of one divided image but cannot be detected by object detection for one divided image to be detected by object detection for the other divided image. Appropriate numbers can be set as the first number and the second number. The types of divided images are not limited to two types, the first number of first divided images and the second number of second divided images. In this embodiment, a decrease in the accuracy of object detection in the boundary portion of one of the first and second divided images can be compensated for by the object detection result for the other divided image. In another embodiment of the present invention, the image division means divides the original image into at least a first odd number of first divided images and a first even number of second divided images. Typically, the original image is divided into the same number of divisions (odd or even) at equal intervals in the vertical and horizontal directions. The types of divided images are not limited to two types: a first odd number of first divided images and a first even number of second divided images. In this embodiment, the first and second split images can be split images with approximately the same aspect ratio as that of the original image. Therefore, even if object detection processing is performed on the split images using image processing means used in current object detection devices, there is no distortion due to changes in image scale, and object detection performance is not affected. In another embodiment of the present invention, the image dividing means divides the original image into a plurality of divided image groups each including at least one type of divided image and differing in the total number of divided images. The number, shape, etc. of the divided images constituting each divided image group can be appropriately set. Image Processing Means teeth, If the object is not included in the object detection result for the original image, the selected Based on divided image information indicating each divided image constituting one divided image group, For each original image that composes one selected split image group, compose one selected split image group. Detect objects contained in each divided image Perform object detection processing And, Compose one selected split image group If the object is included in any of the object detection results for each divided image, Compose one selected split image group Object detection results for each divided image Output .on the other hand, Compose one selected split image group If the object detection result for each divided image does not include an object, the same process is performed for a different divided image group. The repetition of the object detection process for the different divided image groups can be terminated at an appropriate timing. For example, the process can be terminated when the number of divided image groups for which the object detection process has been performed reaches a set value or when a set time has elapsed since the start of the object detection process. This embodiment can be preferably used in the case where the presence of at least one object in an original image is detected. In this embodiment, since the object processing can be terminated at the point in time when the presence of an object in the original image is detected, the processing load on the object detection means can be reduced. In a different embodiment of the present invention, the object detection means includes a detection result synthesis means. When the image processing means outputs an object detection result for the original image, the detection result synthesis means outputs the object detection result for the original image as the object detection result for the original image (the object detection result for the original image based on the object detection result for the original image), and when the image processing means outputs an object detection result for each divided image, the detection result synthesis means synthesizes the object detection results for each divided image and outputs it as the object detection result for the original image (the object detection result for the original image based on the object detection result for each divided image). As a method for synthesizing the object detection results for each divided image, for example, a method is used in which the position information of the object in the divided image is converted into the position information in the original image. In addition, when multiple object detection results include objects of the same category and position, for example, a method is used in which the object with the higher object-likeness score used in the object detection process is selected. This embodiment can be preferably used in the case where the presence of at least one object in an original image is detected. This embodiment can reduce the processing load on the object detection means. In a different embodiment of the present invention, the object detection means includes a detection result synthesis means. If the object detection result for the original image does not include an object, the image processing means performs object detection processing for each divided image based on divided image information indicating each divided image to detect objects contained in each divided image, and outputs the object detection result for each divided image and the object detection result for the original image. The detection result synthesis means includes: When the image processing means outputs the object detection results for each divided image and the object detection results for the original image, The object detection result for each divided image is combined with the object detection result for the original image, and output as the object detection result for the original image (the object detection result for the original image based on the object detection result for each divided image and the object detection result for the original image). As a method for combining the object detection result for each divided image with the object detection result for the original image, for example, a method of converting the position information of the object in the divided image into the position information in the original image can be used. In addition, when multiple object detection results include objects of the same category and position, for example, a method of selecting the object with the higher score indicating the object-likeliness used in the object detection process can be used. In this embodiment, it is possible to further improve the object detection accuracy. can. Effect of the Invention

[0006] The present invention can improve the object detection accuracy easily and inexpensively. [Brief description of the drawings]

[0007] [Figure 1] 1 is a block diagram of an embodiment of an object detection device of the present invention. [Diagram 2] FIG. 1 is a diagram showing an overview of an example of an image processing means for detecting an object using deep learning. [Diagram 3] 11A and 11B are diagrams illustrating detection accuracy when an object is detected based on divided images and when an object is detected based on an original image. [Figure 4]FIG. 4 is a diagram illustrating a first example of the operation of the object detection device according to the embodiment. [Diagram 5] FIG. 11 is a diagram illustrating a second example of the operation of the object detection device according to an embodiment. [Figure 6] FIG. 11 is a diagram illustrating a third example of the operation of the object detection device according to the embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, an embodiment of an object detection device of the present invention will be described with reference to the drawings. FIG. 1 shows a block diagram of an object detection device according to an embodiment. The object detection device of one embodiment is composed of a processing means 10, an imaging means 20, a storage means 30, an input means 40, an output means 50, etc., which are connected via a wired communication line, a wireless communication line, or the like. The imaging means 20 is configured by, for example, a digital camera using a CCD or CMOS. The imaging means 20 outputs image information indicating the captured image. The imaging means 20 corresponds to the "imaging means for capturing an original image" of the present invention, and the image information output from the imaging means 20 corresponds to the "image information indicating the original image" of the present invention. The storage means 30 is configured with a ROM, a RAM, etc., and stores programs for executing the processes of the processing means 10 and various data. The input means 40 is composed of a keyboard, a touch panel, etc., and is used to input various information. The output means 50 is composed of a display means such as a liquid crystal display device or an organic EL display device, a printing means, etc., and outputs various information. Note that when a display means that allows information to be input by touching a display portion displayed on a display screen is used as the display means, the input means 40 can be omitted. The imaging means 20, the storage means 30, the output means 50, etc. may be located at a location separate from the processing means 10.

[0009] The processing means 10 is composed of a CPU and the like. The processing means 10 has an object detection means 11. The object detection means 11 has an image processing means 11a, an image division means 11b, and a detection result synthesis means 11c.

[0010] The image division means 11b divides a captured image (hereinafter referred to as "original image") indicated by image information output from the imaging means 20 into a plurality of divided images. The image information includes image information output from the imaging means 20 and stored in the storage means 30. A method for dividing the original image by the image division means 11b will be described later.

[0011] The image processing means 11a detects the category and position of an object included in a divided image based on divided image information indicating the divided images divided by the image dividing means 11b. The image processing means 11a can also detect the category and position of an object included in an original image based on image information indicating the original image. The image processing means 11a may be any of various known image processing means that detects the category and position of an object contained in an original image or a divided image based on image information indicating an original image or divided image information indicating a divided image using deep learning. For example, an image processing means that detects an object using a method such as SSD or YOLO may be used. For example, as shown in Fig. 2, SSD is based on a multi-layered CNN (Convolutional Neural Network) and is composed of a layer that estimates object existence region candidates and a layer that determines the category of the object in the existence region candidate. In the layer that estimates object existence region candidates, image information is divided into multiple rectangular regions (default boxes) of a predetermined size, and object existence region candidates (bounding boxes) are estimated while taking into account the deviation of the rectangular regions. In the layer that determines the category of the object in the existence region candidate, a separately trained CNN is used to determine the category of the object in the existence region candidate.

[0012] The detection result synthesis means 11c synthesizes the object detection results for each divided image by the image processing means 11a, and outputs the result as the object detection result for the original image. For example, the object detection results for each divided image are synthesized after converting the position information of the object in each divided image into position information in the original image, and output as the object detection result for the original image (in this case, the object detection result for the original image based on the object detection results for each divided image). The detection result synthesis means 11c can synthesize the object detection results for each divided image and the object detection results for the original image by the image processing means 11a, and output it as the object detection result for the original image (in this case, the object detection result for the original image based on the object detection results for each divided original image and the object detection result for the original image). At this time, if the object detection results to be synthesized include objects with almost the same category and position, for example, the one with the higher score indicating the object-likeliness used during object detection is selected. Alternatively, both can be output. Furthermore, the detection result synthesis means 11c can also output the object detection result for the original image as the object detection result for the original image (in this case, "the object detection result for the original image based on the object detection result for the original image").

[0013] First, with reference to FIG. 3, we will explain the object detection accuracy when using current image processing means as the image processing means and performing object detection processing on an original image, and when performing object detection processing on divided images obtained by dividing the original image. M1 shows an image of a doll placed at a distance. M2 shows an image obtained by reducing the size of the doll image M1 to a size corresponding to the original image used in the object detection process. Note that the size of the image M2 in the original image changes when the zoom function of the imaging means 20 is used. M3 shows an image obtained by reducing the size of the doll image M1 to correspond to the divided images obtained by dividing the original image (in FIG. 3, four divided images obtained by dividing the image vertically and horizontally into two equal parts at equal intervals). N1 indicates an image (processing target image) of a region in the original image that corresponds to the reduced doll image M2. N2 indicates an image (processing target image) of a region in the divided image that corresponds to the reduced doll image M3. When object detection processing was performed on the processing target image N1 of the original image using the current image processing means, the doll image M2 could not be detected. On the other hand, when object detection processing was performed on the processing target image N2 of the divided image, the doll image M3 could be detected. In this way, by performing object detection processing by the image processing means on the divided images obtained by dividing the original image, it becomes possible to detect objects that cannot be detected in the original image. In other words, by dividing the original image into divided images and performing object detection processing on the divided images, it is possible to improve the accuracy of object detection. In the experiment, the minimum detectable size of an object in the original image was approximately [17 x 40 pixels], while the minimum detectable size of an object in the 4-part image was approximately [6 x 12 pixels].

[0014] Next, the operation of the object detection device of this embodiment will be described. A first example of the operation of the object detection device of this embodiment will be described with reference to FIG. In the first embodiment, the image dividing means 11b divides the original image into four (two evenly spaced horizontally × two evenly spaced vertically) four-part images (a)-(d) by four-part division lines (one horizontal division line, one vertical division line). The positions of each of the four-part images (a)-(d) in the original image (for example, the positions of the corners of each of the four-part images on the coordinate system of the original image) are stored in the storage means 30. In the first embodiment, the original image is divided into 2 squares (2 2 ) (two equally spaced vertically and horizontally) into four divided images (a)-(d). In other words, the aspect ratios of the four divided images (a)-(d) are equal to (including "almost equal") the aspect ratio of the original image. For this reason, when the image processing means 11a performs object detection processing on each of the four divided images (a)-(d), there is no distortion due to a change in image scale, and there is no impact on object detection performance.

[0015] The image processing means 11a executes object detection processing for each of the four divided images (a)-(d) and outputs the object detection results for each of the four divided images (a)-(d). Also, image processing means 11a performs an object detection process on the original image and outputs the object detection result for the original image. The detection result synthesis means 11c synthesizes the object detection results for each of the four divided images (a) to (d) and the object detection results for the original image, and outputs them as the object detection results for the original image (the object detection results for each divided image and the object detection results for the original image based on the object detection results for the original image). For example, the object detection results for each of the four divided images (a) to (d) are synthesized into object position information in the original image, and the object detection results for each of the four divided images (a) to (d) and the original image are output. To The object detection results for the corresponding images are synthesized.

[0016] In the first embodiment, object detection processing is performed on four quarter images (a) to (d) obtained by dividing the original image into four parts, thereby making it possible to detect objects that cannot be detected by object detection processing on the original image. This makes it possible to improve the accuracy of object detection. In addition, when an original image is divided into four quarter-divided images (a) to (d), it may be impossible to detect an object that exists in the boundary portion between each of the quarter-divided images (a) to (d), for example, an object that exists across the boundary portion between the quarter-divided images (a) to (d). For example, as shown in Fig. 5, an object (P) that exists across the quarter-divided images (a) and (b) may not be detected by object detection processing on the quarter-divided images (a) and (b). In the first embodiment, an object detection process is executed on the original image, and an object detection result for the original image is output. Then, the object detection result for the original image and the object detection result for each of the four divided images (a)-(d) are combined. This makes it possible to detect objects that exist in the boundary portions of each of the four divided images (a)-(d), which cannot be detected by the object detection process for each of the four divided images (a)-(d), by the object detection process for the original image.

[0017] A second example of the operation of the object detection device of this embodiment will be described with reference to FIG. In the second embodiment, the image division means 11b divides the original image into four 4-part images (a)-(d) by 4-part division lines, and into nine 9-part images (A)-(I) by 9-part division lines (3 horizontal division lines, 3 vertical division lines). The positions of each of the 4-part images (a)-(d) and each of the 9-part images (A)-(I) in the original image (for example, the positions of the corners of each divided image on the coordinate system of the original image) are stored in the storage means 30. In the second embodiment, the original image is divided into 2 squares (2 2 ) (two equally spaced vertically and horizontally) in four divided images (a) to (d), and 2 ) (three equally spaced in both the vertical and horizontal directions) into nine divided images (A)-(I). In other words, the aspect ratios of the four divided images (a)-(d) and the nine divided images (A)-(I) are equal (including "almost equal") to the aspect ratio of the original image. For this reason, when the image processing means 11a performs object detection processing on the four divided images (a)-(d) and the nine divided images (A)-(I), there is no distortion due to a change in image scale, and there is no impact on object detection performance.

[0018] The image processing means 11a performs object detection processing on each of the four-division images (a)-(d) and each of the nine-division images (A)-(I), and outputs object detection results for each of the four-division images (a)-(d) and each of the nine-division images (A)-(I). Moreover, the image processing means 11a executes an object detection process on the original image and outputs the object detection result for the original image. The detection result synthesis means 11c synthesizes the object detection results for each of the four-division images (a)-(d) and each of the nine-division images (A)-(I) with the object detection result for the original image, and outputs it as the object detection result for the original image (the object detection result for the original image based on the object detection result for each of the divided images and the object detection result for the original image). For example, the object detection results for each of the four-division images (a)-(d) and each of the nine-division images (A)-(I) are synthesized with the object detection result for the original image in a state where the object position information included in the object detection results for each of the four-division images (a)-(d) and each of the nine-division images (A)-(I) is converted to position information in the original image. The synthesis process of the object detection results by the detection result synthesis means 11c can use a method similar to the synthesis process in the first embodiment.

[0019] In the second embodiment, object detection processing is performed on four 4-division images (a) to (d) obtained by dividing an original image into four parts and nine 9-division images (A) to (I) obtained by dividing an original image into nine parts. Object detection processing It can detect objects that cannot be detected by conventional methods. This makes it possible to improve the accuracy of object detection. In addition, the original image is divided into two parts, each of which is an even number (2 2 ) and divide it into four parts, each part being an odd number equal to the square of 3 (3 2 As a result, the boundaries (vertical and horizontal boundaries) of the four-part images (a) to (d) intersect with the boundaries (vertical and horizontal boundaries) of the nine-part images (A) to (I), but do not overlap or run parallel to each other. Therefore, the decrease in object detection accuracy at the boundary portion of each of the four-division images (a) to (d) (for example, an object existing across the boundary portion cannot be detected) can be compensated for by the object detection results of each of the nine-division images (A) to (I). For example, as shown in FIG. 5, an object (P) existing across the four-division images (a) and (b) may not be detectable by the object detection process for the four-division images (a) and (b), but it can be detected by the object detection process for the nine-division image (B). Similarly, the decrease in object detection accuracy at the boundary portion of each of the nine-division images (A) to (I) can be compensated for by the object detection results of each of the four-division images (a) to (d). Furthermore, it can also be compensated for by the object detection process for the original image. Therefore, the object detection accuracy can be further improved.

[0020] A third example of the operation of the object detection device of this embodiment will be described with reference to FIG. In the third embodiment, the image division means 11b divides the original image into four 4-division images (a)-(d) by 4-division lines, nine 9-division images (A)-(I) by 9-division lines, and 16 (4 equally spaced horizontally × 4 equally spaced vertically) 16-division images (1)-(16) by 16-division lines (4 horizontal division lines, 4 vertical division lines). The positions of the 4-division images (a)-(d), the 9-division images (A)-(I), and the 16-division images (1)-(16) in the original image (for example, the positions of the corners of each divided image on the coordinate system of the original image) are stored in the storage means 30. In the third embodiment, the original image is divided into 2 squares (2 2 ) (two equally spaced vertically and horizontally) in four divided images (a) to (d), and 2 ) (3 equally spaced vertically and horizontally) in 9 divided images (A) to (I) and 4 squared (4 2) (four equally spaced vertically and horizontally). That is, the aspect ratios of the four-division images (a) to (d), the nine-division images (A) to (I), and the sixteen-division images (1) to (16) are equal (including "almost equal") to the aspect ratio of the original image. For this reason, when the image processing means 11a performs object detection processing on the four-division images (a) to (d), the nine-division images (A) to (I), and the sixteen-division images (1) to (16), there is no distortion due to a change in image scale, and there is no impact on object detection performance.

[0021] The image processing means 11a performs object detection processing on each of the four-division images (a)-(d), each of the four-division images (A)-(I), and each of the sixteen-division images (1)-(16), and outputs object detection results for each of the four-division images (a)-(d), each of the nine-division images (A)-(I), and each of the sixteen-division images (1)-(16). Moreover, the image processing means (11a) executes an object detection process on the original image and outputs the object detection result for the original image. The detection result synthesis means 11c synthesizes the object detection results for each of the 4-division images (a)-(d), the object detection results for each of the 9-division images (A)-(I), and the object detection results for each of the 16-division images (1)-(16) with the object detection result for the original image, and outputs it as the object detection result for the original image (the object detection result for the original image based on the object detection results for each of the division images and the object detection result for the original image). The synthesis process of the object detection results by the detection result synthesis means 11c can use the same method as the synthesis process in the first and second embodiments.

[0022] In the third embodiment, object detection processing is performed on four 4-division images (a) to (d) obtained by dividing the original image into four parts, nine 9-division images (A) to (I) obtained by dividing the original image into nine parts, and sixteen 16-division images (1) to (16) obtained by dividing the original image into sixteen parts, thereby making it possible to detect objects that cannot be detected by object detection processing performed on the original image. This makes it possible to improve the accuracy of object detection. In addition, the original image is divided into two parts, each of which is an even number (2 2) into 4 parts and 4 squared parts (4 2 ) and divide it into 16 parts, each of which is an odd number equal to the square of 3 (3 2 As a result, parts of the boundaries between the 4-division images (a)-(d) and the 16-division images (1)-(16) overlap in parallel, but the boundaries between the 4-division images (a)-(d) and the 16-division images (1)-(16) do not overlap in parallel with the boundaries between the 9-division images (A)-(I). Therefore, the decrease in object detection accuracy at the boundary parts of each of the 9-division images (A)-(I) (for example, an object that exists across the boundary parts cannot be detected) can be compensated for by the object detection results of each of the 4-division images (a)-(d) and each of the 16-division images (1)-(16). Similarly, the decrease in object detection accuracy at the boundary parts of each of the 4-division images (a)-(d) and each of the 16-division images (1)-(16) can be compensated for by the object detection results of each of the 9-division images (A)-(I). Furthermore, it can also be compensated for by the object detection results of the original image. Therefore, the object detection accuracy can be further improved.

[0023] In the above embodiment, the image processing means 11a performs object detection processing on each divided image and also performs object detection processing on the original image, and the detection result synthesis means 11c synthesizes the object detection results for each divided image and the object detection results for the original image to detect the category and position of the object contained in the original image, but the present invention can also be configured as follows. As a divided image, the original image is divided into 2 squared (2 2 ) (equally spaced in two vertically and horizontally), the original image is divided into 4 divided images, 2 ) divided into 4 parts and the square of 3 (3 2 A group of divided images, including 9 divided images divided into 3 equal intervals vertically and horizontally, 2 ) divided into 4 parts, 2 ) divided into 9 parts and the square of 4 (4 2) pieces (4 equal intervals vertically and horizontally), the divided images or combinations of divided images that make up the divided image group are not limited to this. Although the object detection results for each divided image and the object detection results for the original image are combined to output the object discrimination result for the original image, it is also possible to configure the object discrimination results for the divided images to be combined to output the object discrimination result for the original image. Although one divided image group is used, multiple divided image groups are used, and the object discrimination process is performed on the divided images constituting one selected divided image group, and if the object discrimination result does not include an object, a different divided image group is selected, and the object discrimination process is performed on the divided images constituting the selected divided image group. The repetition of the object discrimination process on different divided image groups can be terminated at an appropriate timing. For example, the process can be terminated when the number of divided image groups on which the object detection process has been performed reaches a set value or when a set time has elapsed since the start of the object detection process. When the purpose is to detect the presence or absence of an object (that at least one object is present), the following configuration can be used. The image processing means 11a executes an object detection process for the original image, and executes an object detection process for each divided image when the object detection result for the original image does not include an object. The detection result synthesis means 11c outputs the object detection result for the original image as the object detection result for the original image when the object detection result for the original image includes an object, and synthesizes the object detection results for each divided image and outputs it as the object detection result for the original image when the object detection result for the original image does not include an object. Note that, when the object detection result for each divided image does not include an object, it is also possible to configure the apparatus to execute object detection processes for a different number of divided images, synthesize the object detection results for the different number of divided images, and output them as the object detection result for the original image. The repetition of the object discrimination process for each divided image of a different number can be ended, for example, at the same timing as the timing of ending the repetition of the object discrimination process for the different divided image groups described above. In this case, the number of times the object detection process is performed by the image processing means 11a can be reduced.

[0024] The present invention is not limited to the configurations described in the embodiments, and various modifications, additions, and deletions are possible. The image dividing means, image processing means, and detection result combining means are not limited to the configurations described in the embodiment. The configuration of the object detection means is not limited to the configuration described in the embodiment. Each of the configurations described in the embodiments can be used alone or in combination with an appropriately selected plurality of configurations. [Explanation of symbols]

[0025] 10 Processing means 11 Object detection means 11a Image processing means 11b Image segmentation means 11c Processing result processing means 20 Imaging means 30 Memory means 40 Input Method 50 Output Means

Claims

1. The apparatus includes an imaging means and an object detection means, The imaging means outputs image information indicating an original image captured, The object detection means includes an image processing means and an image division means, The image dividing means divides an original image indicated by the image information output from the imaging means into a plurality of divided images; The object detection device is characterized in that the image processing means performs an object detection process on the original image based on image information indicating the original image to detect an object contained in the original image, and if the object detection result for the original image includes an object, outputs the object detection result for the original image, and if the object detection result for the original image does not include an object, performs an object detection process on each of the divided images based on divided image information indicating each of the divided images to detect an object contained in the divided image, and outputs the object detection result for each divided image.

2. The object detection device according to claim 1 , the image dividing means divides the original image into at least a first number of first divided images and a second number of second divided images; An object detection device characterized in that the first number and the second number are set so that the boundary portions of the first divided image and the boundary portions of the second divided image are parallel and do not overlap.

3. The object detection device according to claim 1 or 2, The object detection device, characterized in that the image division means divides the original image at least into first divided images that are the square of a first odd number and into second divided images that are the square of a first even number.

4. The object detection device according to any one of claims 1 to 3, the image dividing means is capable of dividing the original image into a plurality of divided image groups each including at least one type of divided image and differing in total number of divided images; The object detection device is characterized in that, when an object is not included in the object detection result for the original image, the image processing means performs object detection processing for each divided original image that constitutes the selected one divided image group based on divided image information indicating each divided image that constitutes the selected one divided image group, to detect an object included in each divided image that constitutes the selected one divided image group, and when an object is included in any of the object detection results for each divided image that constitutes the selected one divided image group, the image processing means outputs the object detection result for each divided image that constitutes the selected one divided image group, and when an object is not included in the object detection result for each divided image that constitutes the selected one divided image group, the image processing means performs similar processing on a different divided image group.

5. The object detection device according to any one of claims 1 to 4, The object detection means has a detection result synthesis means, The object detection device is characterized in that, when an object detection result for an original image is output from the image processing means, the detection result synthesis means outputs the object detection result for the original image as the object detection result for the original image, and, when an object detection result for each divided image is output from the image processing means, the detection result synthesis means synthesizes the object detection results for each divided image and outputs it as the object detection result for the original image.

6. The object detection device according to any one of claims 1 to 3, The object detection means has a detection result synthesis means, When an object is not included in the object detection result for the original image, the image processing means executes an object detection process for each of the divided images based on divided image information indicating each of the divided images to detect an object included in the divided images, and outputs the object detection result for each of the divided images and the object detection result for the original image; The object detection device is characterized in that, when the image processing means outputs an object detection result for each divided image and an object detection result for the original image, the detection result synthesis means synthesizes the object detection result for each divided image with the object detection result for the original image and outputs it as the object detection result for the original image.

Images