Object detection device, object detection method, and storage medium

Abstract

An object detection apparatus, object detection method, and storage medium are provided that can maintain high detection accuracy while reducing processing load. The object detection apparatus includes: an acquisition unit that acquires an image captured during an image capture process; a low-resolution image generation unit that generates a low-resolution image that reduces the image quality of the captured image; a definition unit that defines one or more partial region groups, each of which is defined by cutting out multiple partial regions from the low-resolution image; and an extraction unit that, for each of the partial regions included in the one or more partial region groups, derives a summary value obtained by summarizing the differences in feature quantities between the summarizing values ​​and those of surrounding partial regions, and extracts a region of interest based on the summary value.

Description

Technical Field

[0001] This invention relates to an object detection device, an object detection method, and a storage medium. Background Technology

[0002] Previously, an invention of a driving obstacle detection system was disclosed. This system divides the area of ​​an object in a monitored area such as a road obtained by taking pictures into blocks, extracts local feature quantities for each block, and determines the presence or absence of an obstacle based on the extracted local feature quantities (Patent Document 1).

[0003] Prior art literature

[0004] Patent documents

[0005] Patent Document 1: Japanese Patent Application Publication No. 2019-124986 Summary of the Invention

[0006] The problem that the invention aims to solve

[0007] Previous technologies have been prone to issues such as excessive processing load or insufficient accuracy.

[0008] The present invention was made with this situation in mind, and one of its objectives is to provide an object detection device, object detection method, and storage medium that can perform object detection while reducing the processing load.

[0009] Solution for solving the problem

[0010] The object detection device, object detection method, and storage medium of the present invention adopt the following structure.

[0011] (1): An object detection apparatus according to one aspect of the present invention includes: an acquisition unit that acquires a captured image obtained on a shooting path; a low-resolution image generation unit that generates a low-resolution image that reduces the image quality of the captured image; a definition unit that defines one or more partial region groups, each of the one or more partial region groups being defined by cutting out multiple partial regions from the low-resolution image; and an extraction unit that, for each of the one or more partial region groups, derives a summary value obtained by summarizing the differences in feature quantities between the partial regions and the surrounding partial regions, and extracts a region of interest based on the summary value.

[0012] (2): In the above (1) scheme, the defining unit defines multiple partial region groups in such a way that the number of pixels in the partial region is different from each other among the multiple partial region groups, and the extraction unit extracts the region of interest by performing a pixel-wise addition operation on the aggregated value among the multiple partial region groups.

[0013] (3): In the above scheme (1) or (2), the extraction unit summarizes the differences in feature quantities between the partial regions included in the one or more partial region groups and other partial regions that are adjacent to each other in the upper, lower, left, right and diagonal directions to derive the summary value.

[0014] (4): In the above scheme (3), the extraction unit further adds the differences in feature values ​​between vertically adjacent partial regions, the differences in feature values ​​between horizontally adjacent partial regions, and the differences in feature values ​​between diagonally adjacent partial regions to the summary value for the partial regions included in the one or more partial region groups respectively.

[0015] (5): In any of the above schemes (1) to (4), the object detection device further includes a high-resolution processing unit, which performs high-resolution processing on the part of interest in the captured image and determines whether the object on the road is a moving object that should be avoided from contact.

[0016] (6): Another aspect of the present invention provides an object detection method performed using a computer, wherein the object detection method includes: acquiring a captured image obtained on the shooting route; generating a low-resolution image that reduces the image quality of the captured image; defining one or more partial region groups; and for each partial region group, deriving a summary value obtained by summarizing the differences in feature quantities between the summarizing and surrounding partial regions, and extracting a region of interest based on the summary value, wherein each of the one or more partial region groups is defined by cutting out multiple partial regions from the low-resolution image.

[0017] (7): In another embodiment of the present invention, the storage medium stores a program that causes a computer to perform the following processing: acquiring a captured image obtained on the shooting route; generating a low-resolution image that reduces the image quality of the captured image; defining one or more partial region groups; and for each partial region included in the one or more partial region groups, deriving a summary value obtained by summarizing the differences in feature quantities between the summarizing and the surrounding partial regions, and extracting the part of interest based on the summary value, wherein the one or more partial region groups are defined by cutting out multiple partial regions from the low-resolution image.

[0018] Invention Effects

[0019] According to the methods described in (1) to (7) above, object detection can be performed while reducing the processing load.

[0020] Furthermore, based on the method described in (2) above, it is possible to improve the robustness of the detection performance relative to the deviation of the size of the object reflected at the region of interest. Attached Figure Description

[0021] Figure 1 This is a diagram showing an example of the structure and peripheral equipment of the object detection device 100.

[0022] Figure 2 This is a diagram schematically illustrating the functions of the various parts of the object detection device 100.

[0023] Figure 3 This diagram illustrates the processing of the masking area determination unit 130, the mesh definition unit 140, and the extraction unit 150.

[0024] Figure 4 This diagram illustrates the processing of the feature quantity difference calculation unit 152, the summarization unit 154, and the first addition operation unit 156.

[0025] Figure 5 This is a diagram representing a definition example of the surrounding grid.

[0026] Figure 6 This is a diagram illustrating an example of the rules for selecting the target mesh and the source mesh for comparison.

[0027] Figure 7 This is another example of a rule for selecting the target mesh and the source mesh for comparison.

[0028] Figure 8 This diagram illustrates the processing of the first addition unit 156 and the second addition unit 158.

[0029] Explanation of reference numerals in the attached figures:

[0030] 10 cameras

[0031] 100 Object Detection Device

[0032] 110 Obtained by the Department

[0033] 120 Low-resolution image generation unit

[0034] 130 Covert Area Decision Department

[0035] 140 Mesh Definition Section

[0036] 150 Extraction Section

[0037] 152 Characteristic Quantity Difference Calculation Department

[0038] 154 Summary Department

[0039] 156 First Addition Operation Section

[0040] 158 Second Addition Operation Section

[0041] 160 Focus Area Extraction Section

[0042] 170 High-resolution processing unit. Detailed Implementation

[0043] Hereinafter, embodiments of the object detection device, object detection method, and storage medium of the present invention will be described with reference to the accompanying drawings. The object detection device is, for example, mounted on a mobile body. The mobile body refers to, for example, a four-wheeled vehicle, a two-wheeled vehicle, a micro-motorized vehicle, a self-moving robot, or a portable device such as a smartphone mounted on a self-moving mobile body or moved by a person. In the following description, the mobile body is a four-wheeled vehicle, and will be referred to as a "vehicle." The object detection device is not limited to a device mounted on a mobile body; it can also be a device that processes images captured by a fixed-point observation camera or a smartphone camera, and the processing described below is based on such images.

[0044] Figure 1 This diagram illustrates an example of the structure and peripheral equipment of the object detection device 100. The object detection device 100 communicates with the camera 10, the driving control device 200, the reporting device 210, and the like.

[0045] The camera 10 is mounted on the back of the vehicle's windshield, etc., and captures images of at least the road in the direction of the vehicle's travel, then outputs the captured images to the object detection device 100. It should be noted that a sensor fusion device or similar device may also exist between the camera 10 and the object detection device 100, but its description is omitted.

[0046] The driving control device 200 is, for example, an automatic driving control device that enables the vehicle to drive autonomously, a driving support device that performs inter-vehicle distance control, automatic braking control, automatic lane change control, etc. The reporting device 210 is a speaker, vibrator, light source, display device, etc., used to output information to the occupants of the vehicle.

[0047] The object detection device 100 includes, for example, an acquisition unit 110, a low-resolution image generation unit 120, a mesh definition unit 140, an extraction unit 150, and a high-resolution processing unit 170. The extraction unit 150 includes a feature difference calculation unit 152, a summarization unit 154, a first addition operation unit 156, a second addition operation unit 158, and a region of interest extraction unit 160. These components are implemented, for example, by executing a program (software) using a hardware processor such as a CPU (Central Processing Unit). Some or all of these components can be implemented by hardware (including circuitry) such as LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), and GPU (Graphics Processing Unit), or through a combination of software and hardware. The program can be pre-saved on storage devices such as HDD (Hard Disk Drive) and flash memory (storage devices with non-transitory storage media), or it can be saved on removable storage media such as DVD and CD-ROM (non-transitory storage media), and installed by assembling the storage media onto the drive device.

[0048] Figure 2 This diagram schematically illustrates the functions of the various parts of the object detection device 100. Referring also to the following... Figure 2 The various parts of the object detection device 100 will be described. The acquisition unit 110 acquires images from the camera 10. The acquisition unit 110 stores the acquired images (data) in a working memory such as RAM (Random Access Memory).

[0049] The low-resolution image generation unit 120 performs thinning processing on the captured image to generate a low-resolution image with lower image quality than the captured image. A low-resolution image is, for example, an image with fewer pixels than the captured image.

[0050] The shielding area determination unit 130 determines the shielding area that is not considered as the processing target for the structure below the mesh definition unit 140. Details are described later.

[0051] The grid definition unit 140 defines multiple partial region groups in the low-resolution image. Each partial region group is defined by cutting out multiple partial regions (hereinafter referred to as grids) from the low-resolution image. The partial regions are set without gaps, for example, in a rectangular shape. The grid definition unit 140 defines the multiple partial region groups in such a way that the number of pixels within the grid differs from each other among the multiple partial region groups. Hereinafter, the multiple partial region groups are sometimes referred to as first partial region group PA1, second partial region group PA2, ..., kth partial region group PAk. Detailed functions of the grid definition unit 140 will be described later.

[0052] The extraction unit 150 extracts a summary value obtained by summarizing the differences in feature quantities between the extraction unit and the surrounding grids for each of the multiple partial region groups. It then performs an addition operation on the summary value across the multiple partial region groups to extract the region of interest (in the figure, the region that is discontinuous with its surroundings). Detailed functions of each part of the extraction unit 150 are described later.

[0053] The high-resolution processing unit 170 cuts out the portion corresponding to the area of ​​interest in the captured image (simultaneous cutting in the figure) and performs high-resolution processing on it to determine whether an object on the road is one that the vehicle should avoid contact with. For example, the high-resolution processing unit 170 uses a learned model that identifies road signs (an example of objects that the vehicle should avoid contact with) and fallen objects (an example of objects that the vehicle should avoid contact with) from the image to determine whether the image reflected at the area of ​​interest is a road sign, a fallen object, or something unclear (an unlearned object). At this time, the high-resolution processing unit 170 can also further narrow down the area of ​​interest in the captured image to the portion identified as equivalent to a road sign or a fallen object for processing.

[0054] Figure 3 This diagram illustrates the processing of the masking region determination unit 130, the mesh definition unit 140, and the extraction unit 150. For example, the masking region determination unit 130 extracts left-right edge points in a low-resolution image, connecting these linearly arranged edge points to detect the positions of road markings, shoulders, etc. (white lines, driving lane boundaries) in the image. Furthermore, it detects the vehicle's driving lane as the region enclosed by left and right road markings, containing a center point related to the left-right direction of the image. Next, the masking region determination unit 130 determines the portion outside the vehicle's driving lane (the portion above the vanishing point where the road markings intersect on the far side, and the portion further to the left and right ends of the road markings) as the masking region. The mesh definition unit 140 and the extraction unit 150 process the process by excluding the masking regions.

[0055] The grid definition unit 140 defines the first region group PA1 with the largest number of pixels, the second region group PA2 with the second largest number of pixels, and so on, until the kth region group Pak with the fewest pixels. "Definition" refers to determining the grid boundaries for a low-resolution image.

[0056] The following describes the processing of the feature quantity difference calculation unit 152, the summarization unit 154, and the first addition operation unit 156. Figures 4-7 The processing of these functional units is performed by first selecting a partial region group and then selecting the grids of interest one by one within that selected partial region group. Then, if all grids in the selected partial region group are selected as grids of interest and the processing ends, the next partial region group is selected and the same processing is performed. If the processing of all partial region groups is completed, the second addition unit 158 ​​performs an addition operation on the summary value (the second summary value V2 described later) between the partial region groups to generate extraction target data PT as an image, which is then passed to the region of interest extraction unit 160.

[0057] Figure 4 This diagram illustrates the processing of the feature difference calculation unit 152, the summarization unit 154, and the first addition unit 156. The feature difference calculation unit 152 calculates the difference in feature values ​​for each pixel of the target mesh and the source mesh. Feature values ​​are, for example, the brightness values ​​of each component of R, G, and B, with each group of R, G, and B representing a pixel. The target mesh and the source mesh are selected from the mesh of interest and surrounding meshes. Figure 5 This diagram illustrates an example of the definition of a peripheral grid. As shown, grids 2 to 9 that are diagonally adjacent to the grid of interest in the top, bottom, left, and right directions are defined as peripheral grids. The selection method for peripheral grids (peripheral areas) is not limited to this; grids in the top, bottom, left, and right directions can be selected as peripheral grids, or other rules can be followed.

[0058] Compare the target mesh and the source mesh, for example, from Figure 6 Choose from the combinations shown in sequence. Figure 6This diagram illustrates an example of the rules for selecting the target and source comparison grids. The target comparison grid is the grid of interest, and the source comparison grids are selected sequentially from grids 2 to 9. The relationship between the target and source comparison grids can be reversed. Then, the summarization unit 154 calculates the sum of the differences in feature values ​​for each pixel, divides it by the number of pixels n within the grid, and calculates the first summary value V1. The first summary value V1 is replaced with zero and output when the grid of interest corresponds to a masking region. That is, the feature difference calculation unit 152, the summarization unit 154, and the first addition operation unit 156 perform the following processes in parallel or sequentially: processing with grid 1 as the comparison target grid and grid 3 as the comparison source grid; processing with grid 1 as the comparison target grid and grid 8 as the comparison source grid; processing with grid 1 as the comparison target grid and grid 5 as the comparison source grid; processing with grid 1 as the comparison target grid and grid 6 as the comparison source grid; processing with grid 1 as the comparison target grid and grid 2 as the comparison source grid; processing with grid 1 as the comparison target grid and grid 4 as the comparison source grid; processing with grid 1 as the comparison target grid and grid 7 as the comparison source grid; and processing with grid 1 as the comparison target grid and grid 9 as the comparison source grid.

[0059] Comparing the target mesh and the comparison source mesh can be done from Figure 7 Choose from the combinations shown in sequence. Figure 7 This is another example of the rules for selecting the target and source meshes for comparison. The combination of the target and source meshes is not limited to the combination of the mesh of interest and the surrounding meshes, but can also include the combination of the surrounding meshes with each other (in particular, the combination of the top and bottom meshes, the left and right meshes, the top-left and bottom-right meshes, and the top-right and bottom-left meshes).

[0060] More specifically, the method for calculating the difference of feature quantities will be explained. For example, consider the following modes 1 to 4 as methods for calculating the difference of feature quantities. In the following explanation, i represents the identification number of the pixel in the target grid and the source grid (i = 1 to k; k is the number of pixels in the target grid and the source grid, respectively).

[0061] (Mode 1)

[0062] The feature quantity difference calculation unit 152 calculates, for example, the difference in brightness of the R component ΔRi, the difference in brightness of the G component ΔGi, and the difference in brightness of the B component ΔBi (as described above, i = 1 to k) between pixels at the same position in both the target grid and the source grid. Then, for each pixel, it calculates the pixel feature quantity Ppi = ΔRi. 2 +ΔGi 2 +ΔBi 2The maximum or average value of each pixel feature quantity Ppi is used as the difference between the feature quantities of the target grid and the source grid.

[0063] (Mode 2)

[0064] The feature difference calculation unit 152 calculates, for example, the statistical values ​​(called average, median, mode, etc.) Raa of the brightness of the R component of each pixel in the target grid, the statistical value (same) Gaa of the brightness of the G component, and the statistical value (same) Baa of the brightness of the B component, and calculates and compares the statistical values ​​(same) Rab of the brightness of the R component of each pixel in the source grid, the statistical value (same) Gab of the brightness of the G component, and the statistical value (same) Bab of the brightness of the B component, and finds their differences ΔRa (=Raa-Rab), ΔGa (=Gaa-Gab), and ΔBa (=Baa-Bab). Then, it calculates ΔRa as the sum of squares of the brightness differences. 2 +ΔGa 2 +ΔBa2 、 Or the maximum value of the square of the difference in brightness, Max(ΔRa) 2 ΔGa 2 ΔBa 2 The difference between the target mesh and the source mesh is used as the feature quantity difference between them.

[0065] (Mode 3)

[0066] For example, the feature difference calculation unit 152 calculates, for each pixel i in the comparison target grid, a first index value W1ai (=(RB) / (R+G+B)) obtained by dividing the difference in brightness between the R component and the B component by the sum of the brightness of the R, G, and B components, and a second index value W2ai (=(RG) / (R+G+B)) obtained by dividing the difference in brightness between the R component and the G component by the sum of the brightness of the R, G, and B components. Similarly, for each pixel i in the comparison source grid, the feature difference calculation unit 152 calculates, for example, a first index value W1bi (=(RB) / (R+G+B)) obtained by dividing the difference in brightness between the R component and the B component by the sum of the brightness of the R, G, and B components, and a second index value W2bi (=(RG) / (R-+-G+B)) obtained by dividing the difference in brightness between the R component and the G component by the sum of the brightness of the R, G, and B components. Next, the feature difference calculation unit 152 calculates the feature value Ppi = (W1ai - W1bi) for each pixel. 2 +(W2ai-W2bi) 2Then, the feature difference calculation unit 152 calculates the maximum or average value of the feature quantity Ppi of each pixel as the difference between the feature quantities of the comparison target grid and the comparison source grid. It should be noted that by combining the first index value and the second index value, the balance of RGB components in each pixel can be represented. Based on the same idea as above, for example, the brightness of each component of R, G, and B can also be defined as the magnitude of a vector offset by 120 degrees, and the vector sum can be used in the same way as the combination of the first index value and the second index value.

[0067] (Mode 4)

[0068] The feature difference calculation unit 152 calculates, for example, the statistical values ​​of the brightness of the R component (same) Raa, the statistical values ​​of the brightness of the G component (same) Gaa, and the statistical values ​​of the brightness of the B component (same) Baa of each pixel in the target grid, and calculates and compares the statistical values ​​of the brightness of the R component (same) Rab, the statistical values ​​of the brightness of the G component (same) Gab, and the statistical values ​​of the brightness of the B component (same) Bab of each pixel in the source grid. Next, the feature difference calculation unit 152 calculates, for the comparison target grid, a third index value W3a (=(Raa-Baa) / (Raa+Gaa+Baa)) obtained by dividing the difference between the brightness statistics of the R component Raa and the brightness statistics of the B component Baa by the sum of the brightness statistics of the R, G, and B components, and a fourth index value W4a (=(Raa-Gaa) / (Raa+Gaa+Baa)) obtained by dividing the difference between the brightness statistics of the R component Raa and the brightness statistics of the G component Gaa by the sum of the brightness statistics of the R, G, and B components. Similarly, the feature difference calculation unit 152 calculates, for the comparison source grid, a third index value W3b (=(Rab-Bab) / (Rab+Gab+Bab)) obtained by dividing the difference between the brightness statistics of the R component Rab and the brightness statistics of the B component Bab by the sum of the brightness statistics of the R, G, and B components, and a fourth index value W4b (=(Rab-Gab) / (Rab+Gab+Bab)) obtained by dividing the difference between the brightness statistics of the R component Rab and the brightness statistics of the G component Gab by the sum of the brightness statistics of the R, G, and B components. Then, the feature difference calculation unit 152 calculates the difference ΔW3 between the third index value W3a of the comparison target grid and the third index value W3b of the comparison source grid, and the difference ΔW4 between the fourth index value W4a of the comparison target grid and the fourth index value W4b of the comparison source grid, and calculates ΔW3 as the sum of their squares. 2 +ΔW4 2 The maximum value of the square, or Max(ΔW3). 2 ΔW4 2 The difference between the characteristic quantities of the target mesh and the source mesh is used for calculation.

[0069] It should be noted that if the image of the object being processed is a black and white image, the feature difference calculation unit 152 can calculate the difference in brightness values ​​as the difference in feature values ​​between the target grid and the source grid. Alternatively, if the image of the object being processed is an RGB image, the RGB image can be converted to a black and white image, and the difference in orbital values ​​can be calculated as the difference in feature values ​​between the target grid and the source grid.

[0070] return Figure 4 The first addition unit 156 performs an addition operation on the first summary value V1 obtained corresponding to the grid of interest to calculate the second summary value V2. The second summary value V2 is an example of "summary value" in the technical solution. If the process of obtaining the second summary value V2 while changing the grid of interest is completed, data with the second summary value V2 set for all grids is generated for each partial region group.

[0071] If the second addition operation unit 158 ​​generates data that sets the second summary value V2 for all grids, it performs addition operations between partial region groups to generate extraction object data PT. Figure 8 This diagram illustrates the processing of the first addition unit 156 and the second addition unit 158. In the diagram, the smallest rectangle represents one pixel of the low-resolution image. For simplicity, it is assumed that there are three regions: a first region group PA1, a second region group PA2, and a third region group PA3, whose sizes are much smaller than actual dimensions. Furthermore, the second summary value V2 is normalized at any stage to be a value between zero and 1. In the illustrated example, the first region group PA1 is a set of first grids consisting of 16 pixels, the second region group PA2 is a set of second grids consisting of 9 pixels, and the third region group PA3 is a set of third grids consisting of 4 pixels. The second summary value V2 is set for grids that bundle multiple pixels. Therefore, the second addition unit 158 ​​expands the second summary value V2 set in the grid across all pixels within the grid, and performs addition operations on the pixel values ​​of pixels within the region groups to generate the extraction target data PT. It should be noted that there may also be partial regions where pixels correspond one-to-one with the grid. In this case, it is not necessary to expand the second summary value V2 for each pixel. It should also be noted that although examples of grid sizes of 16 pixels, 9 pixels, and 4 pixels are shown, when the method for calculating the difference as a feature quantity uses Mode 2 or Mode 4 as described above, setting the number of pixels on one side to a power of 2, such as 4 pixels, 16 pixels, and 64 pixels, can reduce the computational load of the statistical values.

[0072] The focus area extraction unit 160 extracts the bounding rectangle of the portion of the data PT containing the sum of the aggregated values ​​that is above a threshold, and defines it as the focus area.

[0073] As described above, the high-resolution processing unit 170 performs high-resolution processing on the area obtained by applying only the position of the part of interest to the captured image, and determines whether the object on the road is an object that the vehicle should avoid contact with.

[0074] The high-resolution processing unit 170 outputs the discrimination result to the driving control device 200 and / or the reporting device 210. The driving control device 200 performs automatic braking control, automatic steering control, etc., to prevent the object identified as a "falling object" (actually a region on the image) from contacting the vehicle. The reporting device 210 detects the time-to-market (TTC) between the object identified as a "falling object" (as above) and the vehicle. o When the Collision is less than the threshold, an alarm is output using various methods.

[0075] According to the embodiments described above, by including: an acquisition unit 110 that acquires an image obtained on at least the road in the direction of travel of the photographing vehicle; a low-resolution image generation unit 120 that generates a low-resolution image that reduces the image quality of the photographed image; a grid definition unit 140 that defines one or more partial region groups; and an extraction unit 150 that, for each partial region included in one or more partial region groups, derives a summary value obtained by summarizing the differences in feature quantities between the summary and the surrounding partial regions, and extracts the region of interest based on the summary value, it is possible to maintain high detection accuracy while reducing the processing load.

[0076] If the feature difference calculation unit 152 and the summarization unit 154 are directly applied to the captured image, the processing load increases if the number of pixels increases, raising concerns that the driving control device 200 and the reporting device 210 may not be able to react in time to the approaching falling object. In this regard, the object detection device 100 of the embodiment performs object detection by processing the image after generating a low-resolution image, thereby reducing the processing load.

[0077] Furthermore, according to the embodiment, the grid definition unit 140 defines multiple partial region groups in such a way that the number of pixels within the grids of the multiple partial region groups is different from each other, and the extraction unit 150 extracts the region of interest by performing pixel-wise summation operations on the summation values ​​among the multiple partial region groups. Therefore, the robustness of the detection performance against deviations in the size of the falling object can be improved. This is because, when only low-resolution images are processed, the image quality may be reduced to a level where the presence of the falling object cannot be detected. However, according to the embodiment, through the above-described research, it can be expected that the falling object will exhibit a characteristic quantity in a grid of any size. Based on the above, the object detection apparatus 100 according to the embodiment can maintain high detection accuracy while reducing the processing load.

[0078] The implementation methods described above can be performed as follows.

[0079] The object detection device is configured to include:

[0080] Storage device, which stores a program; and

[0081] Hardware processor,

[0082] The hardware processor executes the program stored in the storage device to perform the following processing:

[0083] To obtain images captured during the shooting process;

[0084] Generate a low-resolution image that reduces the image quality of the captured image;

[0085] Define one or more partial region groups; and

[0086] For each of the aforementioned partial regions within a group of partial regions, a summary value is derived by summarizing the differences in feature quantities between the summary value and the surrounding partial regions. Based on this summary value, the region of interest is extracted.

[0087] The one or more partial region groups are defined by cutting out multiple partial regions from the low-resolution image.

[0088] The above description illustrates specific embodiments of the present invention, but the present invention is not limited to such embodiments in any way, and various modifications and substitutions can be made without departing from the spirit of the present invention.

Claims

1. An object detection device, wherein, The object detection device includes: The acquisition unit obtains the images captured during the shooting process; A low-resolution image generation unit generates a low-resolution image that reduces the image quality of the captured image. A definition unit that defines one or more partial region groups, each of which is defined by cutting out multiple partial regions from the low-resolution image; as well as The extraction unit, for each of the more than one partial region group, derives a summary value obtained by summarizing the differences in feature quantities between the summarizing values ​​and the surrounding partial regions, and extracts the region of interest based on the summary value. The defining part defines multiple partial region groups in such a way that the number of pixels within each partial region group is different from that within the multiple partial region groups. The moving body can move autonomously based on the determination result of the object at the region of interest in the captured image.

2. The object detection device according to claim 1, wherein, The extraction unit extracts the region of interest by performing pixel-wise addition on the aggregated value among multiple groups of said partial regions.

3. The object detection device according to claim 1 or 2, wherein, The extraction unit summarizes the differences in feature quantities between the partial regions included in the one or more partial region groups and other partial regions that are adjacent vertically, horizontally, and diagonally, and derives the summarized value.

4. The object detection device according to claim 3, wherein, The extraction unit further adds the differences in feature values ​​between vertically adjacent partial regions, the differences in feature values ​​between horizontally adjacent partial regions, and the differences in feature values ​​between diagonally adjacent partial regions to the summary value for each of the more than one partial region group.

5. The object detection device according to claim 1 or 2, wherein, The object detection device also includes a high-resolution processing unit, which performs high-resolution processing on the region of interest in the captured image to determine whether an object on the road is a moving object that should be avoided.

6. An object detection method, wherein the object detection method is performed using a computer, wherein, The object detection method includes: To obtain images captured during the shooting process; Generate a low-resolution image that reduces the image quality of the captured image; Define one or more partial region groups; and For each of the aforementioned partial region groups, a summary value is derived by summarizing the differences in feature quantities between the summarizing regions and their surrounding partial regions. Based on this summary value, the region of interest is extracted. The one or more partial region groups are defined by cutting out multiple partial regions from the low-resolution image. The number of pixels within each of the multiple groups of said partial regions differs from that of the other partial regions. The moving body can move autonomously based on the determination result of the object at the region of interest in the captured image.

7. A storage medium, wherein, The storage medium stores a program that causes the computer to perform the following processes: To obtain images along at least the path of the moving object's direction of travel; Generate a low-resolution image that reduces the image quality of the captured image; Define one or more partial region groups; as well as For each of the aforementioned partial region groups, a summary value is derived by summarizing the differences in feature quantities between the summarizing regions and their surrounding partial regions. Based on this summary value, the region of interest is extracted. The one or more partial region groups are defined by cutting out multiple partial regions from the low-resolution image. The number of pixels within each of the multiple groups of said partial regions differs from that of the other partial regions. The moving body can move autonomously based on the determination result of the object at the region of interest in the captured image.

Images