Object detection device, object detection method and object detection program

Abstract

To provide an object detection device capable of reducing a computational load of inference processing when point cloud data is used.SOLUTION: A point cloud acquisition unit 21 acquires point cloud data obtained by applying light to a target region and receiving reflected light reflected at a reflection point. A difference calculation unit 23 specifies difference data, the difference between the acquired point cloud data and reference point cloud data. A region determination unit 24 determines a region corresponding to a cluster region obtained by clustering the difference data, as a detection region. An object detection unit 25 acquires a detection result by inputting the point cloud data of the detection region into an object detection model.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a technique for detecting an object using point cloud data. [Background technology]

[0002] There is a technology that uses a visible light camera to detect objects such as abandoned property. When using a visible light camera to detect objects, the object cannot be detected properly in an environment where there is insufficient light, such as at night, or in an environment where direct sunlight shines on the visible light camera and causes flare.

[0003] There are optical sensors such as LiDAR that collect point cloud data by emitting light and receiving the light reflected from reflection points. LiDAR stands for Light Detection and Ranging. Optical sensors can obtain point cloud data even in environments with insufficient light, such as at night, and in environments where direct sunlight shines on a visible light camera and causes flare. Therefore, detecting objects based on point cloud data obtained using optical sensors is being considered.

[0004] Patent Document 1 describes a method of comparing background point cloud data stored in a memory with current point cloud data, and identifying the appearance of some kind of object if a difference occurs. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Publication No. 2020-118619 Summary of the Invention [Problem to be solved by the invention]

[0006] Point cloud data obtained using optical sensors has a larger volume of data than image data obtained using visible light cameras. Therefore, when point cloud data is used as input to perform inference processing on a deep learning model that detects specific objects, the inference processing requires a large amount of calculation, resulting in the need for expensive hardware. An object of the present disclosure is to reduce the amount of calculation required for inference processing when point cloud data is used. [Means for solving the problem]

[0007] The object detection device according to the present disclosure comprises: a point cloud acquisition unit that acquires point cloud data obtained by irradiating a target area with light and receiving reflected light reflected at reflection points; a difference specifying unit that specifies difference data that is a difference between the point cloud data acquired by the point cloud acquiring unit and reference point cloud data; a region determination unit that determines, as a detection region, a region corresponding to a cluster region obtained by clustering the difference data identified by the difference identification unit; an object detection unit that inputs the point cloud data for the detection area determined by the area determination unit into an object detection model and acquires a detection result; Equipped with. [Effects of the Invention]

[0008] In this disclosure, an area corresponding to the cluster area obtained by clustering the difference data is determined as the detection area, and the detection area is input to an object detection model to obtain the detection result. This makes it possible to reduce the amount of data input to the object detection model and the amount of calculation required for inference processing when point cloud data is used. [Brief explanation of the drawings]

[0009] [Figure 1] 1 is a configuration diagram of an object detection device 10 according to a first embodiment. [Figure 2] FIG. 3 is an explanatory diagram of a voxel definition 31 according to the first embodiment. [Figure 3] 3 is a flowchart showing the flow of processing performed by the object detection device 10 according to the first embodiment. [Figure 4] FIG. 2 is an explanatory diagram of a voxelization process according to the first embodiment. [Figure 5] FIG. 3 is an explanatory diagram of a difference identification process according to the first embodiment. [Figure 6] FIG. 3 is an explanatory diagram of a first method for determining a detection region 46 according to the first embodiment. [Figure 7] FIG. 3 is an explanatory diagram of a first method for determining a detection region 46 according to the first embodiment. [Figure 8] FIG. 4 is an explanatory diagram of a second method for determining a detection region 46 according to the first embodiment. [Figure 9] FIG. 10 is an explanatory diagram of a difference identification process according to the first modification. [Figure 10] FIG. 10 is an explanatory diagram of a difference identification process according to Modification 2. [Figure 11] FIG. 10 is an explanatory diagram of an object detection process according to the second embodiment. [Figure 12] FIG. 10 is an explanatory diagram of an object detection process according to the third embodiment. [Figure 13] FIG. 13 is an explanatory diagram of an object detection process according to Modification 5. DETAILED DESCRIPTION OF THE INVENTION

[0010] Embodiment 1 ***Configuration Description*** The configuration of an object detection device 10 according to the first embodiment will be described with reference to FIG. The object detection device 10 is a computer. The object detection device 10 includes hardware such as a processor 11, a memory 12, a storage 13, and a communication interface 14. The processor 11 is connected to other hardware via signal lines and controls the other hardware.

[0011] The processor 11 is an IC that performs processing. IC stands for Integrated Circuit. Specific examples of the processor 11 include a CPU, a DSP, and a GPU. CPU stands for Central Processing Unit. DSP stands for Digital Signal Processor. GPU stands for Graphics Processing Unit.

[0012] The memory 12 is a storage device that temporarily stores data. Specific examples of the memory 12 include SRAM and DRAM. SRAM stands for Static Random Access Memory. DRAM stands for Dynamic Random Access Memory.

[0013] The storage 13 is a storage device that stores data. Specific examples of the storage 13 include an SSD or an HDD. SSD stands for Solid State Drive. HDD stands for Hard Disk Drive. The storage 13 may also be a portable recording medium such as an SD (registered trademark) memory card, CompactFlash (registered trademark), NAND flash, a flexible disk, an optical disk, a compact disk, a Blu-ray (registered trademark) disk, or a DVD. SD stands for Secure Digital. DVD stands for Digital Versatile Disk.

[0014] The communication interface 14 is an interface for communicating with external devices. Specific examples of the communication interface 14 include Ethernet (registered trademark), USB, and HDMI (registered trademark) ports. USB stands for Universal Serial Bus. HDMI stands for High-Definition Multimedia Interface.

[0015] The object detection device 10 is connected to an optical sensor 40 via a communication interface 14. The optical sensor 40 is a device that acquires point cloud data 41 by irradiating a target area with irradiation light and receiving reflected light reflected at reflection points. The position of each point in the point cloud data 41 is identified by the irradiation angle of the irradiation light and the time from when the irradiation light is emitted until the reflected light is received.

[0016] The object detection device 10 includes, as functional components, a point cloud acquisition unit 21, a voxelization unit 22, a difference identification unit 23, a region determination unit 24, and an object detection unit 25. The functions of each functional component of the object detection device 10 are realized by software. The storage 13 stores a program that realizes the function of each functional component of the object detection device 10. The program is loaded into the memory 12 by the processor 11 and executed by the processor 11. In this way, the function of each functional component of the object detection device 10 is realized.

[0017] The storage 13 also stores a voxel definition 31 .

[0018] 1 shows only one processor 11. However, there may be a plurality of processors 11, and the plurality of processors 11 may cooperate to execute programs that realize the respective functions.

[0019] ***Explanation of Operation*** The operation of the object detection device 10 according to the first embodiment will be described with reference to FIGS. The operation procedure of object detection device 10 according to embodiment 1 corresponds to the object detection method according to embodiment 1. Furthermore, a program that realizes the operation of object detection device 10 according to embodiment 1 corresponds to the object detection program according to embodiment 1.

[0020] The voxel definition 31 according to the first embodiment will be described with reference to FIG. The voxel definition 31 is data that defines the voxels 32. The voxels 32 are three-dimensional regions obtained by dividing the target region 33. The size of the voxels 32 can be set arbitrarily depending on the position in the target region 33. In the first embodiment, the larger the distance from the optical sensor 40, the larger the size of the defined voxels 32.

[0021] The flow of processing performed by the object detection device 10 according to the first embodiment will be described with reference to FIG. (Step S11: Point cloud acquisition process) The point cloud acquisition unit 21 acquires point cloud data 41 obtained by irradiating the target area 33 with light using an optical sensor 40 and receiving light reflected at reflection points.

[0022] (Step S12: Voxelization process) The voxelization unit 22 generates voxel data 42 from the point cloud data 41 acquired in step S11. This will be described in detail with reference to Fig. 4. The voxelization unit 22 sets each point of the point cloud data 41 acquired in step S11 as a target point. The voxelization unit 22 associates the target point with a voxel 32 that includes the position of the target point among the multiple voxels 32 obtained by dividing the target region 33. In this way, voxel data 42 corresponding to the point cloud data 41 is generated.

[0023] (Step S13: Difference Identification Process) The difference identification unit 23 identifies difference data 44, which is the difference between the point cloud data acquired in step S11 and the reference point cloud data. Here, the difference identification unit 23 generates the difference data 44 by comparing the voxelized point cloud data. The data obtained by voxelizing the reference point cloud data is called reference data 43. In other words, the difference identification unit 23 identifies difference data 44, which is the difference between the voxel data 42 and the reference data 43. In the first embodiment, the reference data 43 is voxel data 42 generated from the background point cloud data. That is, the reference data 43 is voxel data 42 generated by performing the processing of step S12 on the background point cloud data. The background point cloud data is point cloud data 41 acquired by the optical sensor 40 in a state where no target object or the like is present in the target region 33.

[0024] Specifically, the difference identification unit 23 sets each of the multiple voxels 32 as a target voxel 32. The difference identification unit 23 determines whether or not there is a corresponding point for the target voxel 32, that is, whether or not there is a difference between the voxel data 42 and the reference data 43, based on the presence or absence of a point in the voxel 32. If there is a difference, the difference identification unit 23 identifies the voxel 32 as difference data 44. 5, when a target voxel 32 has no points associated with it in the reference data 43 but has points associated with it in the voxel data 42, the difference identification unit 23 identifies it as difference data 44. In other words, the difference identification unit 23 does not identify a voxel 32 with points associated with it in the reference data 43 as difference data 44 even if the number of associated points is different.

[0025] (Step S14: Clustering process) The difference identification unit 23 clusters the difference data 44 identified in step S13 to form cluster regions 45. The clustering method may use an existing clustering technique, such as grouping adjacent difference data 44 into the same cluster.

[0026] (Step S15: Area determination process) The region determination unit 24 determines a region corresponding to the cluster region 45 obtained in step S14 as the detection region 46. Here, one or more detection regions 46 are set. The method for determining the detection region 46 will be described later.

[0027] (Step S16: Object detection process) The object detection unit 25 sets each of the one or more detection areas 46 determined in step S15 as a target detection area 46. The object detection unit 25 inputs the point cloud data 41 for the target detection area 46 into an object detection model, and acquires a detection result that is an inference result. That is, the object detection unit 25 inputs the point cloud data 41 for the detection area 46, out of the point cloud data 41 acquired in step S21, into the object detection model. Then, the object detection unit 25 acquires a detection result that is a result of inference performed by the object detection model. The object detection unit 25 may input the difference data 44 for the detection region 46, out of the difference data 44 identified in step S13, into the object detection model.

[0028] (Step S17: Result integration process) The object detection unit 25 generates result data by integrating the detection results for each of the one or more detection areas 46 acquired in step S16, and then outputs the result data.

[0029] The following describes the area determination process (step S15 in FIG. 3) according to the first embodiment. Here, two methods, Method 1 and Method 2, will be described as methods for determining the detection area 46.

[0030] Method 1 will be described with reference to FIG. The region determination unit 24 determines an expanded region 47 that is obtained by expanding the cluster region 45 by the reference range around the cluster region 45 as a detection region 46 . Here, the reference range may be set uniformly in all axial directions, or may be set separately for each axial direction. For example, the reference range may be set to 6 in the x-axis direction, 8 in the y-axis direction, and 10 in the z-axis direction. In this case, if the size of cluster region 45 is 4 in the x-axis direction, 5 in the y-axis direction, and 6 in the z-axis direction, the detection region 46 is expanded by 6 in the x-axis direction to a width of 10 in the x-axis direction, by 8 in the y-axis direction to a width of 13 in the y-axis direction, and by 10 in the z-axis direction to a width of 16 in the y-axis direction. In this case, the region is expanded by 3 on each side in the x-axis direction, by 4 on each side in the y-axis direction, and by 5 on each side in the z-axis direction. Furthermore, the reference range may be set so that it is determined for each axis direction according to the size of the cluster region 45 in that axis direction. For example, the reference range may be set to twice the size of the cluster region 45. In this case, if the size of the cluster region 45 in the x-axis direction is 5, it is expanded by 10 in the x-axis direction, and an area with a width of 15 in the x-axis direction becomes the detection region 46. In this case, the cluster region 45 is expanded by 5 on each side in the x-axis direction to become the detection region 46. The size in the y-axis and z-axis directions is also doubled to match the axial directions.

[0031] 7, it is assumed that there are multiple cluster regions 45. It is also assumed that extended regions 47, which are expanded by a reference range around the multiple cluster regions 45, partially overlap each other. In this case, the region determination unit 24 may determine, as the detection region 46, a merged region obtained by merging the extended regions 47 for the multiple cluster regions 45. In other words, the region determination unit 24 combines the extended regions 47 and determines, as the detection region 46, an area including the extended regions 47.

[0032] Method 2 will be described with reference to FIG. The region determination unit 24 determines, as the detection region 46, the divided region 34 that includes the reference position of the cluster region 45 from among the plurality of divided regions 34 obtained by dividing the target region 33 based on the distance from the optical sensor 40. Here, as a specific example, the reference position of cluster region 45 is the center of cluster region 45. For example, in Fig. 8, target region 33 is divided into divided regions 34 of area a to area d. The center of cluster region 45 is included in area b. Therefore, area b is determined as detection region 46. The reference position of the cluster region 45 may be at least a part of the cluster region 45. In this case, if the cluster region 45 spans multiple divided regions 34, the region formed by joining the multiple divided regions 34 is determined as the detection region 46. 8, the target area 33 is divided horizontally to obtain a plurality of divided areas 34, but the target area 33 may be divided vertically to obtain a plurality of divided areas 34. In this case as well, one or more divided areas 34 that include a cluster area 45 are determined as the detection area 46.

[0033] ***Effects of the First Embodiment*** As described above, the object detection device 10 according to the first embodiment determines an area corresponding to the cluster area 45 obtained by clustering the difference data 44 as the detection area 46, and inputs the detection area 46 into the object detection model to obtain a detection result. This makes it possible to reduce the amount of data input to the object detection model, and to reduce the amount of calculation required for inference processing when the point cloud data 41 is used. In other words, while point cloud data for the entire target area 33 is normally input to the object detection model, only point cloud data for the detection area 46 of the target area 33 is input to the object detection model. This makes it possible to reduce the amount of calculation required for inference processing.

[0034] By reducing the amount of data input to the object detection model, it is possible to reduce the amount of communication and data storage capacity required for input to the object detection model.

[0035] ***Other Configurations*** <Variation 1> In step S13 of FIG. 3, the difference calculation unit 23 may exclude from the difference data 44 the voxels 32 in which the number of associated points in the voxel data is equal to or less than the reference number of points. For example, the number of reference points is set to 2. In this case, as shown in FIG. 9, voxels 32 having two or fewer associated points in the voxel data are excluded from the difference data 44. In FIG. 9, the middle voxel 32 has two associated points and is therefore excluded from the difference data 44. On the other hand, the bottom voxel 32 has three associated points and is therefore left in the difference data 44.

[0036] <Variation 2> In step S14 of FIG. 3, the difference identification unit 23 may exclude from the difference data 44 the voxels 32 that constitute the obtained cluster region 45 in which the number of voxels 32 included in the cluster region 45 is equal to or less than the first voxel number. For example, the first number of voxels is set to 5. In this case, as shown in Fig. 10, the voxels 32 that make up cluster region 45B, which has four voxels 32, are excluded from the difference data 44. On the other hand, the voxels 32 that make up cluster A, which has ten voxels 32, are left as they are in the difference data 44.

[0037] <Variation 3> In the first embodiment, each functional component is realized by software. However, as a third modification, each functional component may be realized by hardware. The following describes the differences between the first embodiment and the third modification.

[0038] When each functional component is realized by hardware, the object detection device 10 includes an electronic circuit 15 instead of the processor 11, the memory 12, and the storage 13. The electronic circuit 15 is a dedicated circuit that realizes the functions of each functional component, the memory 12, and the storage 13.

[0039] The electronic circuit 15 may be a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, a logic IC, a GA, an ASIC, or an FPGA. GA stands for Gate Array. ASIC stands for Application Specific Integrated Circuit. FPGA stands for Field-Programmable Gate Array. Each functional component may be realized by one electronic circuit 15, or each functional component may be realized by distributing it among a plurality of electronic circuits 15.

[0040] <Variation 4> As a fourth modification, some of the functional components may be realized by hardware, and other functional components may be realized by software.

[0041] The processor 11, memory 12, storage 13, and electronic circuit 15 are collectively referred to as a processing circuit. In other words, the functions of the respective functional components are realized by the processing circuit.

[0042] Embodiment 2 The second embodiment differs from the first embodiment in that a different object detection model is used depending on the distance from the optical sensor 40 to the detection area 46. In the second embodiment, this difference will be explained, and explanation of the same points will be omitted.

[0043] ***Explanation of Operation*** The object detection process (step S16 in FIG. 3) according to the second embodiment will be described with reference to FIG. In the second embodiment, a plurality of object detection models are prepared according to the distance from the optical sensor 40. In Fig. 11, an object detection model A corresponding to areas a and b that are close to the optical sensor 40, and an object detection model B corresponding to areas c and d that are far from the optical sensor 40 are prepared. Here, each object detection model is trained using point cloud data 41 of a corresponding distance. Trained using point cloud data 41 of a corresponding distance means trained using point cloud data 41 of a density according to the corresponding distance. That is, in the example of FIG. 11 , object detection model A is trained using point cloud data 41 of areas a and b, which are close to the optical sensor 40. Object detection model B is trained using point cloud data 41 of areas c and d, which are far from the optical sensor 40.

[0044] The object detection unit 25 sets each of the one or more detection regions 46 determined in step S15 as the target detection region 46. The object detection unit 25 inputs the point cloud data 41 for the target detection region 46 into an object detection model corresponding to the distance from the optical sensor 40 to the target detection region 46, and acquires a detection result. In the example of FIG. 11 , if the target detection region 46 is included in area a or area b, the object detection unit 25 inputs the point cloud data 41 for the target detection region 46 to object detection model A. On the other hand, if the target detection region 46 is included in area c or area d, the object detection unit 25 inputs the point cloud data 41 for the target detection region 46 to object detection model B.

[0045] ***Effects of the Second Embodiment*** As described above, the object detection device 10 according to the second embodiment uses different object detection models depending on the distance from the optical sensor 40 to the detection region 46. The density of the point cloud data 41 varies depending on the distance from the optical sensor 40. That is, the closer an area is to the optical sensor 40, the higher the density of the point cloud data 41. Therefore, even for the same object, the appearance of the point cloud data 41 will differ greatly depending on the distance from the optical sensor 40. As a result, when attempting to detect objects over a wide range, the detection accuracy of the object detection model will decrease. For example, it is likely that an object that can be detected in an area close to the optical sensor 40 cannot be detected in an area far from the optical sensor 40. The object detection device 10 according to the second embodiment can improve detection accuracy by using an object detection model according to the distance of the detection area 46 from the optical sensor 40. In particular, each object detection model is trained using point cloud data 41 at the corresponding distance. Therefore, when point cloud data 41 at the corresponding distance is provided as input, the object can be detected with high accuracy.

[0046] Embodiment 3 The third embodiment differs from the first and second embodiments in that whether or not to use an object detection model is controlled depending on the position of the cluster region 45. In the third embodiment, this difference will be described, and a description of the same points will be omitted. In the third embodiment, a case where a modification is made to the second embodiment will be described. However, it is also possible to make modifications to the second embodiment.

[0047] ***Explanation of Operation*** The object detection process (step S16 in FIG. 3) according to the third embodiment will be described with reference to FIG. In the third embodiment, the target area 33 is divided into an area that uses an object detection model and an area that does not use the object detection model. For example, within the target area 33, an area that is close to the optical sensor 40 and is less than a reference distance is considered to be an area that uses the object detection model, and an area that is far from the optical sensor 40 and is the reference distance or more is considered to be an area that does not use the object detection model. In Fig. 12, areas a and b that are close to the optical sensor 40 are areas that use the object detection model, and areas c and d that are far from the optical sensor 40 are areas that do not use the object detection model.

[0048] The object detection unit 25 sets each of the one or more detection regions 46 determined in step S15 as the target detection region 46. The object detection unit 25 determines whether to use an object detection model depending on the position of the cluster region 45 that is the basis of the target detection region 46. When using an object detection model, the object detection unit 25 inputs the point cloud data 41 for the target detection region 46 into the object detection model to obtain a detection result. On the other hand, when not using an object detection model, the object detection unit 25 sets the cluster region 45 that is the basis of the target detection region 46 as the detection result of the target detection region 46.

[0049] ***Effects of the Third Embodiment*** As described above, object detection device 10 according to the third embodiment controls whether to use an object detection model depending on the position of cluster region 45. In particular, object detection device 10 according to the third embodiment does not use an object detection model for regions far from optical sensor 40, and uses cluster region 45 as the detection result as is. As a result, for regions that are far from optical sensor 40 and where it is difficult to detect an object using the object detection model, it is possible to use the portion where an object may exist as the detection result without using the object detection model.

[0050] ***Other Configurations*** <Variation 5> In the third embodiment, a modification is made to the first embodiment. However, it is also possible to make a modification to the second embodiment. A modification to the second embodiment will be described with reference to FIG.

[0051] The target area 33 is divided into three or more areas according to the distance from the optical sensor 40, and the farthest area is set as an area where no object detection model is used. For the remaining areas, an object detection model is prepared for each area. 13, area d, which is the farthest from optical sensor 40, is set as an area where no object detection model is used. Object detection model A, which corresponds to areas a and b, which are close to optical sensor 40, and object detection model B, which corresponds to area c, which is far from optical sensor 40, are prepared.

[0052] The object detection unit 25 sets each of the one or more detection regions 46 determined in step S15 as the target detection region 46. The object detection unit 25 determines whether to use an object detection model depending on the position of the cluster region 45 that is the basis of the target detection region 46. If the object detection model is to be used, the object detection unit 25 inputs the object detection model according to the distance from the optical sensor 40 to the target detection region 46 to obtain the detection result. On the other hand, if the object detection model is not to be used, the object detection unit 25 sets the cluster region 45 that is the basis of the target detection region 46 as the detection result of the target detection region 46.

[0053] In addition, the word "unit" in the above description may be read as a "circuit," "step," "procedure," "process," or "processing circuit."

[0054] Various aspects of the present disclosure are summarized below as appendices. (Appendix 1) a point cloud acquisition unit that acquires point cloud data obtained by irradiating a target area with light and receiving reflected light reflected at reflection points; a difference specifying unit that specifies difference data that is a difference between the point cloud data acquired by the point cloud acquiring unit and reference point cloud data; a region determination unit that determines, as a detection region, a region corresponding to a cluster region obtained by clustering the difference data identified by the difference identification unit; an object detection unit that inputs the point cloud data for the detection area determined by the area determination unit into an object detection model and acquires a detection result; An object detection device comprising: (Appendix 2) The region determination unit determines an extended region obtained by expanding the cluster region by a reference range as the detection region. 10. The object detection device of claim 1. (Appendix 3) When a plurality of cluster regions exist and extended regions expanded by a reference range around the plurality of cluster regions partially overlap each other, the region determination unit determines a merged region obtained by merging the extended regions of the plurality of cluster regions as the detection region. 3. An object detection device as defined in claim 2. (Appendix 4) The region determination unit determines, as the detection region, a divided region that includes a reference position of the cluster region from among a plurality of divided regions obtained by dividing the target region based on a distance from an optical sensor that irradiates the light. 10. The object detection device of claim 1. (Appendix 5) The object detection unit inputs the detection area to an object detection model corresponding to a reference position of the cluster area, among a plurality of object detection models. 5. An object detection device according to any one of claims 1 to 4. (Appendix 6) The plurality of object detection models are trained using point cloud data having a density corresponding to the reference position of the cluster region. 6. The object detection device of claim 5. (Appendix 7) The object detection unit treats a cluster area, among a plurality of divided areas obtained by dividing the area determination unit based on the distance from the optical sensor that irradiates the light, in which the reference position is included in a divided area whose distance from the optical sensor is equal to or greater than a reference distance, as an output of the object detection model. 7. An object detection device according to any one of claims 1 to 6. (Appendix 8) The object detection device further comprises: a voxelization unit that generates voxel data corresponding to the point cloud data by associating each point of the point cloud data acquired by the point cloud acquisition unit with a voxel that includes a position of the target point among a plurality of voxels obtained by dividing the target region, the voxelization unit treating each point of the point cloud data acquired by the point cloud acquisition unit as a target point, The difference identification unit identifies, for each of the plurality of voxels, a difference cell having a difference between the voxel data generated from the point cloud data by the voxel generation unit and the voxel data generated from the reference point cloud data by the voxel generation unit, as the difference data, based on the presence or absence of a corresponding point. 8. An object detection device according to any one of claims 1 to 7. (Appendix 9) The computer acquires point cloud data obtained by irradiating the target area with light and receiving light reflected at reflection points; The computer identifies difference data that is a difference between the acquired point cloud data and reference point cloud data, a computer determines, as a detection region, a region corresponding to a cluster region obtained by clustering the difference data; An object detection method in which a computer inputs the point cloud data for the detection area into an object detection model to obtain a detection result. (Appendix 10) a point cloud acquisition process for acquiring point cloud data obtained by irradiating a target area with light and receiving reflected light reflected at a reflection point; a difference identification process for identifying difference data that is a difference between the point cloud data acquired by the point cloud acquisition process and reference point cloud data; a region determination process for determining, as a detection region, a region corresponding to a cluster region obtained by clustering the difference data identified by the difference identification process; an object detection process in which the point cloud data for the detection area determined by the area determination process is input into an object detection model to obtain a detection result; An object detection program that causes a computer to function as an object detection device that performs the above.

[0055] The embodiments and modifications of the present disclosure have been described above. Some of these embodiments and modifications may be combined and implemented. Also, one or more of them may be implemented partially. Note that the present disclosure is not limited to the above embodiments and modifications, and various modifications are possible as needed. [Explanation of symbols]

[0056] 10 Object detection device, 11 Processor, 12 Memory, 13 Storage, 14 Communication interface, 21 Point cloud acquisition unit, 22 Voxelization unit, 23 Difference identification unit, 24 Area determination unit, 25 Object detection unit, 31 Voxel definition, 32 Voxel, 33 Target area, 34 Segmented area, 40 Optical sensor, 41 Point cloud data, 42 Voxel data, 43 Reference data, 44 Difference data, 45 Cluster area, 46 Detection area, 47 Extended area.

Claims

1. a point cloud acquisition unit that acquires point cloud data obtained by irradiating a target area with light and receiving reflected light reflected at reflection points; a difference specifying unit that specifies difference data that is a difference between the point cloud data acquired by the point cloud acquiring unit and reference point cloud data; a region determination unit that determines, as a detection region, a region corresponding to a cluster region obtained by clustering the difference data identified by the difference identification unit; an object detection unit that inputs the point cloud data for the detection area determined by the area determination unit into an object detection model and acquires a detection result; An object detection device comprising:

2. The region determination unit determines an extended region obtained by expanding the cluster region by a reference range as the detection region. The object detection device according to claim 1 .

3. When a plurality of cluster regions exist and extended regions expanded by a reference range around the plurality of cluster regions partially overlap each other, the region determination unit determines a merged region obtained by merging the extended regions of the plurality of cluster regions as the detection region. The object detection device according to claim 2 .

4. The region determination unit determines, as the detection region, a divided region that includes a reference position of the cluster region from among a plurality of divided regions obtained by dividing the target region based on a distance from an optical sensor that irradiates the light. The object detection device according to claim 1 .

5. The object detection unit inputs the detection area to an object detection model corresponding to a reference position of the cluster area, among a plurality of object detection models. The object detection device according to claim 1 .

6. The plurality of object detection models are trained using point cloud data having a density corresponding to the reference position of the cluster region. The object detection device according to claim 5 .

7. The object detection unit treats a cluster area, among a plurality of divided areas obtained by dividing the area determination unit based on the distance from the optical sensor that irradiates the light, in which the reference position is included in a divided area whose distance from the optical sensor is equal to or greater than a reference distance, as an output of the object detection model. The object detection device according to claim 1 .

8. The object detection device further comprises: a voxelization unit that generates voxel data corresponding to the point cloud data by associating each point of the point cloud data acquired by the point cloud acquisition unit with a voxel that includes the position of the target point among a plurality of voxels obtained by dividing the target area, with the target point being a target point; Equipped with The difference identification unit identifies, for each of the plurality of voxels, a difference cell having a difference between the voxel data generated from the point cloud data by the voxel generation unit and the voxel data generated from the reference point cloud data by the voxel generation unit, as the difference data, based on the presence or absence of a corresponding point. The object detection device according to claim 1 .

9. The computer acquires point cloud data obtained by irradiating the target area with light and receiving light reflected at reflection points; The computer identifies difference data that is a difference between the acquired point cloud data and reference point cloud data, a computer determines, as a detection region, a region corresponding to a cluster region obtained by clustering the difference data; An object detection method in which a computer inputs the point cloud data for the detection area into an object detection model to obtain a detection result.

10. a point cloud acquisition process for acquiring point cloud data obtained by irradiating a target area with light and receiving reflected light reflected at a reflection point; a difference identification process for identifying difference data that is a difference between the point cloud data acquired by the point cloud acquisition process and reference point cloud data; a region determination process for determining, as a detection region, a region corresponding to a cluster region obtained by clustering the difference data identified by the difference identification process; an object detection process in which the point cloud data for the detection area determined by the area determination process is input into an object detection model to obtain a detection result; An object detection program that causes a computer to function as an object detection device that performs the above.

Images