A target detection method and device and a storage medium

Abstract

The application discloses a target detection method and device and a storage medium, relates to the technical field of automatic driving, and can improve the accuracy of obstacle target sensing and the safety of automatic driving. The method comprises the following steps: detecting a first object in first point cloud data by using a target detection model to obtain a first bounding box and a category of the first object, and first target point cloud data included in the first bounding box; performing clustering processing on the first point cloud data to obtain a second bounding box and a category of a second object, and second target point cloud data included in the second bounding box; if it is determined that the first object and the second object are the same target object according to the categories of the first object and the second object and an intersection-over-union of the first bounding box and the second bounding box, when the size of the second bounding box is greater than the size of the first bounding box, third target point cloud data for identifying the target object is determined from the first target point cloud data and the second target point cloud data, and the target object is identified.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and in particular to a target detection method, device, and storage medium. Background Technology

[0002] Obstacle perception technology is a key technology in the field of autonomous driving. It is used to perceive obstacles encountered by the vehicle during operation, track the obstacles, and obtain the relative position between the obstacle and the vehicle. Based on the relative position between the obstacle and the vehicle, and using obstacle avoidance strategies, the vehicle can automatically avoid obstacles, thereby achieving autonomous driving.

[0003] In related technologies, obstacles are directly perceived and tracked based on data collected by LiDAR installed on vehicles. However, this method can result in missed or false detections of obstacles, leading to low accuracy in obstacle detection. Summary of the Invention

[0004] This invention provides a target detection method, apparatus, and storage medium that can improve the accuracy of obstacle target perception, thereby enhancing the safety of autonomous driving.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] In a first aspect, the present invention provides a target detection method, the method comprising:

[0007] A target detection model is used to detect the first object included in the first point cloud data, and a first bounding box representing the location of the first object, the category of the first object, and the first target point cloud data included in the first bounding box are obtained; the first point cloud data is based on the raw point cloud data collected by the radar on the vehicle in the current acquisition cycle;

[0008] Clustering is performed on the first point cloud data to obtain a second bounding box representing the location of the second object, the category of the second object, and the second target point cloud data included in the second bounding box;

[0009] Based on the categories of the first object and the second object, and the intersection-union ratio of the first bounding box and the second bounding box, determine whether the first object and the second object are the same target object;

[0010] If the first object and the second object are determined to be the same target object, then based on the size relationship between the first bounding box and the second bounding box, a third target point cloud data for identifying the target object is determined from the first target point cloud data and the second target point cloud data.

[0011] The target object is identified based on the third target point cloud data.

[0012] The target detection method provided by this invention processes the first point cloud data obtained from the raw point cloud data collected by the radar on the vehicle during the current acquisition cycle using both a target detection model and clustering methods, resulting in two processing results: a first bounding box, category, and first target point cloud data located within the first bounding box of a first object; and a second bounding box, category, and second target point cloud data located within the second bounding box of a second object. These two results are then correlated and fused. If the first and second objects are determined to be the same target object based on the intersection-union ratio (IUU) of the first and second bounding boxes and their categories, then a third target point cloud data corresponding to the target object is determined based on the first and second target point cloud data, thus obtaining the target object from the third target point cloud data. This target detection method, which correlates and fuses processing results obtained through two different methods, avoids false detections of target objects compared to existing methods that identify target objects using only one method, resulting in higher accuracy. Especially when two objects have similar features, the target detection method provided by this invention achieves higher accuracy in obtaining the target object. Accurately identifying target objects during autonomous driving can improve the safety of autonomous driving.

[0013] In one possible implementation, the above-mentioned clustering process on the first point cloud data yields a second bounding box representing the location of the second object, the category of the second object, and the second target point cloud data included in the second bounding box, including:

[0014] Identify the first feature point in the first point cloud data that belongs to the target plane, where the target plane is the plane on which the ground is located, determined based on the target road information;

[0015] The first feature point is deleted from the first point cloud data to obtain multiple second feature points;

[0016] Clustering is performed on multiple second feature points to obtain the clusters corresponding to the second object and the category labels of the second object;

[0017] Based on the cluster, a second bounding box representing the location of the second object is determined, along with the second target point cloud data included within the second bounding box, and the category of the second object is determined based on the category label.

[0018] In one possible implementation, determining the first feature point belonging to the target plane in the first point cloud data includes:

[0019] Obtain target road information within a preset range for the vehicle;

[0020] Determine the plane equation of the target plane based on the coordinates of the road feature points in the target road information;

[0021] Divide each feature point in the first target point cloud data into multiple grids;

[0022] Based on the feature points in each grid and the plane equation, determine the distance between each grid and the target plane;

[0023] If the distance is less than the first threshold, then each feature point in the grid is determined as the first feature point.

[0024] In one possible implementation, determining the distance between each grid and the target plane based on the feature points in each grid and the plane equation includes:

[0025] The centroid coordinates of each grid are determined based on the coordinates of each feature point in each grid.

[0026] Based on each centroid coordinate and plane equation, determine the distance between each grid and the target plane.

[0027] In one possible implementation, determining whether the first object and the second object are the same target object based on the categories of the first object and the second object, and the intersection-union ratio of the first bounding box and the second bounding box, includes:

[0028] If the first object and the second object are of the same category, and the intersection-union ratio of the first bounding box and the second bounding box is greater than the second threshold, then the first object and the second object are determined to be the same target object.

[0029] If the first object and the second object are of different categories, or if the intersection-union ratio of the first bounding box and the second bounding box is less than the second threshold, then the first object and the second object are determined to be different target objects.

[0030] In one possible implementation, the determination of third target point cloud data for identifying the target object from the first and second target point cloud data based on the size relationship between the first and second bounding boxes includes:

[0031] Determine the size relationship between the first bounding box and the second bounding box;

[0032] If the size of the second bounding box is larger than the size of the first bounding box, then the size of the second bounding box is reduced to the size of the first bounding box to obtain the third bounding box;

[0033] Delete the point cloud data of the second target that is outside the third bounding box to obtain the fourth target point cloud data;

[0034] The point cloud data of the first target and the point cloud data of the fourth target are merged to obtain the point cloud data of the third target.

[0035] In one possible implementation, after determining the size relationship between the first bounding box and the second bounding box, the following is also included:

[0036] If the size of the second bounding box is smaller than the size of the first bounding box, then feature points in the second target point cloud data that do not belong to the first target point cloud data are identified as noise points.

[0037] Noise is removed from the second target point cloud data to obtain the third target point cloud data.

[0038] In one possible implementation, before employing an object detection model to detect the first object included in the first point cloud data, the following is also included:

[0039] Acquire raw point cloud data from the main radar mounted on the roof of the vehicle, as well as raw point cloud data from multiple secondary radars mounted on the vehicle body.

[0040] Motion compensation was performed on the raw point cloud data of the main radar and the raw point cloud data of the secondary radar respectively.

[0041] The motion-compensated original point cloud data of the main radar and the original point cloud data of each secondary radar are stitched together to obtain stitched point cloud data.

[0042] Remove irrelevant feature points located outside the preset range of the vehicle from the stitched point cloud data to obtain the first point cloud data.

[0043] In one possible implementation, before performing motion compensation on the raw point cloud data of the main radar and the raw point cloud data of the secondary radar, the target detection method further includes:

[0044] Determine the first transformation matrix of the vehicle relative to the target radar from the vehicle coordinate system to the map coordinate system, where the target radar is the main radar and each of the multiple secondary radars.

[0045] In one possible implementation, the motion compensation of the raw point cloud data of the main radar includes:

[0046] Obtain the second transformation matrix of the main radar from the map coordinate system to the main radar coordinate system, and the third transformation matrix of the main radar from the main radar coordinate system to the vehicle coordinate system;

[0047] Motion compensation is performed on the raw point cloud data of the main radar based on the first transformation matrix, the second transformation matrix, and the third transformation matrix.

[0048] The above motion compensation for the raw point cloud data of the secondary radar includes:

[0049] Obtain the fourth transformation matrix of the secondary radar from the map coordinate system to the corresponding secondary radar coordinate system, the fifth transformation matrix of the secondary radar from the secondary radar coordinate system to the vehicle coordinate system, and the sixth transformation matrix of the secondary radar from the secondary radar coordinate system to the main radar coordinate system.

[0050] Motion compensation is performed on the raw point cloud data of the secondary radar based on the first transformation matrix, the fourth transformation matrix, the fifth transformation matrix, and the sixth transformation matrix.

[0051] In one possible implementation, the aforementioned preset range is a three-dimensional spatial range. Irrelevant feature points located outside the preset range of the vehicle in the stitched point cloud data are deleted to obtain the first point cloud data, which includes:

[0052] Feature points in the stitched point cloud data that are outside the preset plane range of the vehicle, and feature points whose axis coordinates are outside the preset coordinate range, are identified as irrelevant feature points.

[0053] Remove irrelevant feature points from the stitched point cloud data to obtain the first point cloud data.

[0054] In one possible implementation, after determining whether the first object and the second object are the same target object, the target detection method further includes:

[0055] If it is determined that the first object and the second object are not the same target object, then the second point cloud data, the third point cloud data, and the fourth point cloud data are obtained. The second point cloud data, the third point cloud data, and the fourth point cloud data are obtained based on the raw point cloud data collected by the radar in the three consecutive acquisition cycles after the current acquisition cycle.

[0056] Clustering is performed on the second point cloud data, the third point cloud data, and the fourth point cloud data respectively to obtain the fourth bounding box and the category of the fourth object, the fifth bounding box and the category of the fifth object, and the sixth bounding box and the category of the sixth object.

[0057] The existence of the second object is determined based on the categories of the second, fourth, fifth, and sixth objects, and the intersection-union ratio of every two bounding boxes in the second, fourth, fifth, and sixth bounding boxes.

[0058] Secondly, the present invention provides a target detection device, the target detection device comprising:

[0059] The first determining unit is used to detect the first object included in the first point cloud data using a target detection model, and obtain a first bounding box representing the location of the first object, the category of the first object, and the first target point cloud data included in the first bounding box; the first point cloud data is obtained based on the raw point cloud data collected by the radar on the vehicle in the current acquisition cycle;

[0060] The second determining unit is used to perform clustering processing on the first point cloud data to obtain a second bounding box representing the location of the second object, the category of the second object, and the second target point cloud data included in the second bounding box;

[0061] The third determining unit is used to determine whether the first object and the second object are the same target object based on the categories of the first object and the second object, and the intersection-union ratio of the first bounding box and the second bounding box.

[0062] The third determining unit is further configured to, if the first object and the second object are determined to be the same target object, determine, based on the size relationship between the first bounding box and the second bounding box, a third target point cloud data for identifying the target object from the first target point cloud data and the second target point cloud data;

[0063] The third determining unit is also used to identify the target object based on the third target point cloud data.

[0064] In one possible implementation, the second determining unit described above is specifically used for:

[0065] Identify the first feature point in the first point cloud data that belongs to the target plane, where the target plane is the plane on which the ground is located, determined based on the target road information;

[0066] The first feature point is deleted from the first point cloud data to obtain multiple second feature points;

[0067] Clustering is performed on multiple second feature points to obtain the clusters corresponding to the second object and the category labels of the second object;

[0068] Based on the cluster, a second bounding box representing the location of the second object is determined, along with the second target point cloud data included within the second bounding box, and the category of the second object is determined based on the category label.

[0069] In one possible implementation, the second determining unit described above is specifically used for:

[0070] Obtain target road information within a preset range for the vehicle;

[0071] Determine the plane equation of the target plane based on the coordinates of the road feature points in the target road information;

[0072] Divide each feature point in the first target point cloud data into multiple grids;

[0073] Based on the feature points in each grid and the plane equation, determine the distance between each grid and the target plane;

[0074] If the distance is less than the first threshold, then each feature point in the grid is determined as the first feature point.

[0075] In one possible implementation, the second determining unit described above is specifically used for:

[0076] The centroid coordinates of each grid are determined based on the coordinates of each feature point in each grid.

[0077] Based on each centroid coordinate and plane equation, determine the distance between each grid and the target plane.

[0078] In one possible implementation, the third determining unit described above is specifically used for:

[0079] If the first object and the second object are of the same category, and the intersection-union ratio of the first bounding box and the second bounding box is greater than the second threshold, then the first object and the second object are determined to be the same target object.

[0080] If the first object and the second object are of different categories, or if the intersection-union ratio of the first bounding box and the second bounding box is less than the second threshold, then the first object and the second object are determined to be different target objects.

[0081] In one possible implementation, the third determining unit described above is specifically used for:

[0082] Determine the size relationship between the first bounding box and the second bounding box;

[0083] If the size of the second bounding box is larger than the size of the first bounding box, then the size of the second bounding box is reduced to the size of the first bounding box to obtain the third bounding box;

[0084] Delete the point cloud data of the second target that is outside the third bounding box to obtain the fourth target point cloud data;

[0085] The point cloud data of the first target and the point cloud data of the fourth target are merged to obtain the point cloud data of the third target.

[0086] In one possible implementation, the third determining unit described above is further configured to:

[0087] If the size of the second bounding box is smaller than the size of the first bounding box, then feature points in the second target point cloud data that do not belong to the first target point cloud data are identified as noise points.

[0088] Noise is removed from the second target point cloud data to obtain the third target point cloud data.

[0089] In one possible implementation, the target detection device further includes:

[0090] The acquisition unit is used to acquire the raw point cloud data of the main radar installed on the roof of the vehicle, as well as the raw point cloud data of multiple secondary radars installed on the vehicle body.

[0091] The fourth determining unit is used to perform motion compensation on the original point cloud data of the main radar and the original point cloud data of the secondary radar respectively; to stitch together the motion-compensated original point cloud data of the main radar and the original point cloud data of each secondary radar to obtain stitched point cloud data; and to delete irrelevant feature points located outside the preset range of the vehicle in the stitched point cloud data to obtain the first point cloud data.

[0092] In one possible implementation, the aforementioned fourth determining unit is further configured to determine a first transformation matrix of the vehicle relative to the target radar from the vehicle coordinate system to the map coordinate system, wherein the target radar is the main radar and each of the plurality of secondary radars.

[0093] In one possible implementation, the fourth determining unit described above is specifically used for:

[0094] Obtain the second transformation matrix of the main radar from the map coordinate system to the main radar coordinate system, and the third transformation matrix of the main radar from the main radar coordinate system to the vehicle coordinate system; and perform motion compensation on the original point cloud data of the main radar based on the first transformation matrix, the second transformation matrix, and the third transformation matrix.

[0095] The fourth transformation matrix of the secondary radar from the map coordinate system to the corresponding secondary radar coordinate system, the fifth transformation matrix of the secondary radar from the secondary radar coordinate system to the vehicle coordinate system, and the sixth transformation matrix of the secondary radar from the secondary radar coordinate system to the main radar coordinate system are obtained; and motion compensation is performed on the original point cloud data of the secondary radar based on the first transformation matrix, the fourth transformation matrix, the fifth transformation matrix, and the sixth transformation matrix.

[0096] In one possible implementation, the aforementioned preset range is a three-dimensional spatial range. The fourth determining unit is specifically used for:

[0097] Feature points in the stitched point cloud data that are outside the preset plane range of the vehicle and whose axis coordinates are outside the preset coordinate range are identified as irrelevant feature points; and irrelevant feature points are deleted from the stitched point cloud data to obtain the first point cloud data.

[0098] In one possible implementation, the third determining unit is further configured to acquire second point cloud data, third point cloud data, and fourth point cloud data if it is determined that the first object and the second object are not the same target object. The second point cloud data, third point cloud data, and fourth point cloud data are respectively obtained based on the original point cloud data acquired by the radar in three consecutive acquisition cycles after the current acquisition cycle.

[0099] Clustering is performed on the second point cloud data, the third point cloud data, and the fourth point cloud data respectively to obtain the fourth bounding box and the category of the fourth object, the fifth bounding box and the category of the fifth object, and the sixth bounding box and the category of the sixth object.

[0100] The existence of the second object is determined based on the categories of the second, fourth, fifth, and sixth objects, and the intersection-union ratio of every two bounding boxes in the second, fourth, fifth, and sixth bounding boxes.

[0101] Thirdly, the present invention provides a target detection apparatus, comprising a processor and a memory. The memory stores computer program code, including computer instructions. When the processor executes the computer instructions, the target detection apparatus performs a target detection method as described in the first aspect and any possible implementation thereof.

[0102] Fourthly, the present invention provides a computer-readable storage medium having computer instructions stored thereon, which, when executed on a target detection device, cause the target detection device to perform a target detection method as described in the first aspect or any of the possible implementations of the first aspect. Attached Figure Description

[0103] Figure 1 This is one of the structural schematic diagrams of the target detection device provided in the embodiments of the present invention;

[0104] Figure 2 This is one of the flowcharts illustrating the target detection method provided in an embodiment of the present invention;

[0105] Figure 3 The second schematic flowchart of the target detection method provided in the embodiment of the present invention;

[0106] Figure 4 The third schematic flowchart of the target detection method provided in the embodiments of the present invention;

[0107] Figure 5 The fourth schematic flowchart of the target detection method provided in the embodiments of the present invention;

[0108] Figure 6This is a second schematic diagram of the target detection device provided in an embodiment of the present invention;

[0109] Figure 7 This is the third schematic diagram of the target detection device provided in the embodiment of the present invention. Detailed Implementation

[0110] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0111] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of embodiments of this disclosure, unless otherwise stated, "a plurality of" means two or more. Furthermore, the use of "based on" or "according to" implies openness and inclusiveness, because processes, steps, calculations, or other actions "based on" or "according to" one or more of the stated conditions or values ​​may in practice be based on additional conditions or beyond the stated values.

[0112] To improve the accuracy of obstacle target perception and thus enhance the safety of autonomous driving, this invention provides a target detection method, apparatus, and storage medium. The first point cloud data, obtained from raw point cloud data collected by the vehicle's radar during the current acquisition cycle, is processed using both a target detection model and clustering methods, yielding two results. These two results are then correlated and fused to obtain the target object corresponding to the first point cloud data. This target detection method, which correlates and fuses results obtained through two different methods, avoids false detections and achieves higher accuracy compared to existing methods that identify target objects using only one method. Accurate target object identification during autonomous driving improves the safety of the system.

[0113] The target detection method provided in this invention can be applied to vehicles, which have a main radar mounted on the roof, multiple secondary radars mounted around the vehicle body, an onboard terminal, and a positioning module. The main radar, each secondary radar, and the positioning module establish communication connections via wired or wireless communication.

[0114] The main radar and each secondary radar, as environmental perception devices in the autonomous driving system, collect raw point cloud data in each scanning cycle and send the raw point cloud data to the vehicle terminal.

[0115] The positioning module is used to provide the vehicle terminal with target road information within a preset range of the vehicle, as well as the vehicle's position and pose information.

[0116] Figure 1 This is one of the structural schematic diagrams of a target detection device, such as... Figure 1 As shown, the target detection device may include: a processor 11, a memory 12, a communication interface 13, and a bus 14. The processor 11, the memory 12, and the communication interface 13 can be connected via the communication bus 14.

[0117] The processor 11 is the control center of the target detection device. It can be a single processor 11 or a collective term for multiple processing elements. For example, the processor 11 can be a general-purpose central processing unit (CPU) or other general-purpose processors 11. Among them, the general-purpose processor 11 can be a microprocessor 11 or any conventional processor 11.

[0118] As one embodiment, processor 11 may include one or more CPUs, for example, Figure 1 CPU0 and CPU1 are shown.

[0119] The memory 12 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, a random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a disk storage medium or other magnetic storage device, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto.

[0120] In one possible implementation, the memory 12 can exist independently of the processor 11. The memory 12 can be connected to the processor 11 via a bus 14 and is used to store instructions or program code. When the processor 11 calls and executes the instructions or program code stored in the memory 12, it can implement the target detection method provided in the following embodiments of the present invention.

[0121] In another possible implementation, the memory 12 can also be integrated with the processor 11.

[0122] Communication interface 13 is used for the target detection device to connect with other devices via a communication network, which may be Ethernet, radio access network (RAN), wireless local area network (WLAN), etc. Communication interface 13 may include a receiving unit for receiving data and a transmitting unit for transmitting data.

[0123] Bus 14 can be an Industry Standard Architecture (ISA) bus 14, a Peripheral Component Interconnect (PCI) bus 14, or an Extended Industry Standard Architecture (EISA) bus 14, etc. This bus 14 can be divided into an address bus 14, a data bus 14, a control bus 14, etc. For ease of representation, Figure 1 The bus 14 is represented by a single thick line, but this does not mean that there is only one bus 14 or only one type of bus 14.

[0124] It should be pointed out that, Figure 1 The structure shown does not constitute a limitation on the target detection device, except Figure 1 In addition to the components shown, the target detection device may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0125] The target detection method provided in this embodiment of the invention is executed by a target detection device. This target detection device can be a vehicle, an on-board terminal within the vehicle, or a control module within the vehicle used for target detection. This embodiment of the invention uses a vehicle executing the target detection method as an example to illustrate the target detection method provided by this invention.

[0126] The target detection method provided by the embodiments of the present invention will now be described with reference to the accompanying drawings.

[0127] like Figure 2 As shown, the target detection method provided in this embodiment of the invention includes the following steps 201 to 205.

[0128] 201. Use an object detection model to detect the first object included in the first point cloud data, and obtain the first bounding box representing the location of the first object, the category of the first object, and the first target point cloud data included in the first bounding box.

[0129] The first point cloud data is obtained based on the raw point cloud data collected by the vehicle's radar during the current acquisition cycle. The first object can be obstacles such as pedestrians, roadblocks, or other vehicles around the vehicle. The first bounding box can be a three-dimensional bounding box.

[0130] For example, the target detection model used in this embodiment of the invention is the PointPillar model, but it is not limited to this.

[0131] 202. Perform clustering processing on the first point cloud data to obtain a second bounding box representing the location of the second object, the category of the second object, and the second target point cloud data included in the second bounding box.

[0132] The second object can also be obstacles such as pedestrians, roadblocks, or other vehicles around the vehicle. The second bounding box is also a three-dimensional bounding box.

[0133] 203. Determine whether the first object and the second object are the same target object based on the categories of the first object and the second object, and the intersection-union ratio of the first bounding box and the second bounding box.

[0134] The results of the model detection method and the clustering method on the first point cloud data are combined to determine whether the first object and the second object are the same target object. This can avoid false detections and ensure the accuracy of the target detection results.

[0135] 204. If the first object and the second object are determined to be the same target object, then based on the size relationship between the first bounding box and the second bounding box, a third target point cloud data for identifying the target object is determined from the first target point cloud data and the second target point cloud data.

[0136] In this embodiment of the invention, the second bounding box and the second target point cloud data are obtained by clustering the first point cloud data. During this process, the first point cloud data may be over-segmented or under-segmented. This results in lower accuracy for the second target point cloud data and the second bounding box. Therefore, the accuracy of the results obtained using an object detection model is higher. Based on this, if the first object and the second object are determined to be the same target object, but the size of the second bounding box obtained through clustering differs from the size of the first bounding box obtained using the object detection model, it is necessary to redetermine the third target point cloud data based on the first target point cloud data in the first bounding box and the second target point cloud data in the second bounding box. This prevents false detections and improves the accuracy of target object detection.

[0137] Optionally, determining third target point cloud data for identifying the target object from the first and second target point cloud data based on the size relationship between the first and second bounding boxes may include the following steps: determining the size relationship between the first and second bounding boxes; if the size of the second bounding box is larger than the size of the first bounding box, then reducing the size of the second bounding box to the size of the first bounding box to obtain the third bounding box; deleting point cloud data outside the third bounding box from the second target point cloud data to obtain the fourth target point cloud data; and merging the first and fourth target point cloud data to obtain the third target point cloud data.

[0138] Optionally, if the size of the second bounding box is smaller than the size of the first bounding box, then the first target point cloud data within the first bounding box is used as the standard, and feature points in the second target point cloud data that do not belong to the first target point cloud data are identified as noise. Then, these noise points are deleted from the second target point cloud data to obtain the third target point cloud data.

[0139] 205. Identify the target object based on the third target point cloud data.

[0140] Once the third target point cloud data is determined, conventional technical means can be used to identify the target object based on the third target point cloud data. This embodiment of the invention will not elaborate on this process.

[0141] The target detection method provided in this invention processes the first point cloud data obtained from the raw point cloud data collected by the radar on the vehicle during the current acquisition cycle using both a target detection model and clustering methods, resulting in two processing results: a first bounding box, category, and first target point cloud data located within the first bounding box of a first object; and a second bounding box, category, and second target point cloud data located within the second bounding box of a second object. These two results are then correlated and fused. If the first and second objects are determined to be the same target object based on the intersection-union ratio (IUU) of the first and second bounding boxes and their categories, then a third target point cloud data corresponding to the target object is determined based on the first and second target point cloud data, thereby obtaining the target object from the third target point cloud data. This target detection method, which correlates and fuses processing results obtained through two different methods, avoids false detections of target objects compared to existing methods that identify target objects using only one method, resulting in higher accuracy. Especially when two objects have similar features, the target detection method provided in this invention achieves higher accuracy. Accurately identifying target objects during autonomous driving can improve the safety of autonomous driving.

[0142] Combination Figure 2 ,like Figure 3 As shown, step 202 above may include steps 301 to 304.

[0143] 301. Determine the first feature point in the first point cloud data that belongs to the target plane. The target plane is the plane on which the ground is located, determined based on the target road information.

[0144] Optionally, determining the first feature point may include the following steps: First, acquiring target road information within a preset range of the vehicle. Second, determining the plane equation of the target plane based on the coordinates of the road feature points in the target road information. Third, dividing each feature point in the first target point halo data into multiple grids. Next, determining the distance between each grid and the target plane based on the feature points in each grid and the plane equation. Finally, if the distance is less than a first threshold, then each feature point in the grid is determined as the first feature point.

[0145] For example, the target road information may include lane line information located on the ground, as well as curb information. In determining the target plane equation, the plane equation of the target plane is obtained by fitting the coordinates of multiple road feature points in the target road information.

[0146] For example, determining the distance between each grid cell and the target plane may include the following steps: First, determining the centroid coordinates of each grid cell based on the coordinates of each feature point in each grid cell. Then, determining the distance between each grid cell and the target plane based on each centroid coordinate and the plane equation.

[0147] By using the distance between the centroid coordinates of each grid and the target plane, we can determine whether each feature point in each grid is a point on the target plane. This not only reduces the amount of computation and improves computational efficiency, but also deletes as many points as possible that are located on the target plane, so as to ensure that the remaining second feature points are all feature points that are not on the target plane.

[0148] 302. Delete the first feature point from the first point cloud data to obtain multiple second feature points.

[0149] In the first point cloud data, some feature points are located on the ground, and the point cloud data formed by these ground-based feature points mostly exhibits a textured appearance. Directly performing clustering processing on this first point cloud data would affect the accuracy of the clustering results. To improve the accuracy of the clustering results, feature points located on the target plane containing the ground in the first point cloud data need to be removed before clustering.

[0150] 303. Perform clustering on multiple second feature points to obtain the cluster corresponding to the second object and the category label of the second object.

[0151] In the process of determining the first feature point, the first point cloud data is divided into multiple grids, and then grids whose distance to the target plane is less than a first threshold are removed. That is, the multiple second feature points in the first point cloud data are located in different grids. At this point, adjacent grids need to be determined based on the distances between the multiple grids. Then, the second feature points in adjacent grids are merged, and outliers are removed to obtain a cluster and its corresponding category label. In other words, the cluster corresponding to the second object and its category label are obtained.

[0152] 304. Based on the cluster, determine the second bounding box representing the location of the second object, and the second target point cloud data included in the second bounding box, and determine the category of the second object based on the category label.

[0153] Combination Figure 3 ,like Figure 4 As shown, step 203 above may include step 401 or step 402.

[0154] 401. If the first object and the second object are of the same category, and the intersection-union ratio of the first bounding box and the second bounding box is greater than the second threshold, then the first object and the second object are determined to be the same target object.

[0155] If the categories of the first object and the second object are different, it indicates that the first object and the second object are definitely not the same target object. If the categories of the first object and the second object are the same, it is necessary to determine whether the first object and the second object are the same target object based on the intersection-union ratio (IUR) of the first bounding box and the second bounding box. The larger the IUR of the first bounding box and the second bounding box, the higher the overlap between the first bounding box and the second bounding box, thus indicating a greater probability that the first object and the second object are the same target object.

[0156] 402. If the first object and the second object are of different categories, or the intersection-union ratio of the first bounding box and the second bounding box is less than the second threshold, then the first object and the second object are determined to be different target objects.

[0157] Combination Figure 4 ,like Figure 5 As shown, the target detection method provided in this embodiment of the invention may further include the following steps 501 to 504.

[0158] 501. Acquire the raw point cloud data of the main radar installed on the roof of the vehicle, as well as the raw point cloud data of the multiple secondary radars installed on the vehicle body.

[0159] If a vehicle is equipped with only one main radar, many blind spots will exist. Therefore, in this embodiment of the invention, in addition to the main radar mounted on the roof, multiple blind spot-filling radars, or secondary radars, are also installed around the vehicle body. During vehicle operation, each radar collects raw point cloud data within one acquisition cycle.

[0160] For example, in an embodiment of the present invention, four secondary radars may be included, which are respectively located on the front, rear, left and right sides of the vehicle body.

[0161] Optionally, the vehicle is in motion within a data acquisition cycle, resulting in distortion in the raw point cloud data. To eliminate this distortion, motion compensation is required for the raw point cloud data from the main radar and the secondary radars. Before performing motion compensation on the raw point cloud data from the main radar and the secondary radars, it is necessary to determine the first transformation matrix from the vehicle coordinate system to the map coordinate system relative to the target radar, where the target radar is the main radar and each of the multiple secondary radars.

[0162] For example, taking the target radar as the main radar, the method for determining the first transformation matrix is ​​described. First, based on the start time in the original point cloud data of the main radar, the first pose information of the vehicle at the start time is obtained from the vehicle's positioning module; and based on the end time in the original point cloud data of the main radar, the second pose information of the vehicle at the end time is obtained from the vehicle's positioning module. Then, at the target time, based on the start time, end time, first pose information, and second pose information, the target pose information of the vehicle at the target time is determined. The first transformation matrix can be obtained based on the target pose information.

[0163] For example, the first transformation matrix can be determined according to the following formulas (1) and (2).

[0164]

[0165]

[0166] Among them, Pose time Represents target pose information; time represents target time; begin_time represents start time; end_time represents end time; Pose begin_time Indicates the first pose information; Pose end_time Indicates the second pose information; This represents the first transformation matrix.

[0167] 502. Perform motion compensation on the raw point cloud data of the main radar and the raw point cloud data of the secondary radar respectively.

[0168] Optionally, motion compensation of the raw point cloud data of the main radar may include the following steps: First, obtain the second transformation matrix of the main radar from the map coordinate system to the main radar coordinate system, and the third transformation matrix of the main radar from the main radar coordinate system to the vehicle coordinate system. Then, based on the first, second, and third transformation matrices of the main radar, perform motion compensation on the raw point cloud data of the main radar.

[0169] For example, motion compensation can be performed on the raw point cloud data of the main radar according to the following formula (3).

[0170]

[0171] Among them, X main1 This represents the point cloud data after motion compensation of the original point cloud data from the main radar. Represents the second transformation matrix; Denotes the third transformation matrix; X lidar1 This represents the raw point cloud data of the main radar.

[0172] Optionally, motion compensation of the raw point cloud data of the secondary radar may include the following steps: First, obtain the fourth transformation matrix of the secondary radar from the map coordinate system to the corresponding secondary radar coordinate system, the fifth transformation matrix of the secondary radar from the secondary radar coordinate system to the vehicle coordinate system, and the sixth transformation matrix of the secondary radar from the secondary radar coordinate system to the main radar coordinate system. Then, based on the first, fourth, fifth, and sixth transformation matrices of the secondary radar, perform motion compensation on the raw point cloud data of the secondary radar.

[0173] For example, motion compensation can be performed on the raw point cloud data of the secondary radar according to the following formula (4).

[0174]

[0175] Among them, X main2 This represents the point cloud data after motion compensation of the original point cloud data from the secondary radar. Represents the fourth transformation matrix; Represents the fifth transformation matrix; X represents the sixth transformation matrix; lidar2 This represents the raw point cloud data of the secondary radar.

[0176] It should be noted that the second to sixth transformation matrices can all be obtained directly from the corresponding radar.

[0177] 503. The original point cloud data of the main radar after motion compensation and the original point cloud data of each secondary radar are stitched together to obtain stitched point cloud data.

[0178] For example, taking a vehicle equipped with four secondary radars as an example, the point cloud data of each radar after motion compensation can be added together according to the following formula (5) to obtain the stitched point cloud data.

[0179] X main =X main1 +X main2 +X main3 +X main4 +X main5 (5)

[0180] Among them, X main X represents the splicing of point cloud data. main3 X main4 X main5 These represent the point cloud data after motion compensation for the other three of the four secondary radars.

[0181] 504. Delete irrelevant feature points located outside the preset range of the vehicle in the stitched point cloud data to obtain the first point cloud data.

[0182] To further improve the accuracy of target detection results, it is also necessary to remove irrelevant feature points from the stitched point cloud data.

[0183] Optionally, the aforementioned preset range is a three-dimensional spatial range. Therefore, irrelevant feature points may include feature points in the stitched point cloud data that are outside the preset plane range of the vehicle, and feature points in the stitched point cloud data whose vertical axis coordinates are outside the preset coordinate range. Deleting the aforementioned irrelevant feature points from the stitched point cloud data yields the first point cloud data.

[0184] The preset planar range can be a planar area centered on the vehicle, extending a certain distance in all directions. Deleting irrelevant feature points outside the preset planar range allows us to retain only feature points within the preset planar range, thus enabling the detection of objects only within that range. Radar produces specular reflections, resulting in noise in the point cloud data located in the sky and below the ground. Therefore, it is also necessary to delete point cloud data in the sky and below the ground. When determining the point cloud data in the sky and below the ground, we can use the coordinates of feature points in the stitched point cloud data. After deleting the aforementioned irrelevant feature points from the stitched point cloud data, the resulting first point cloud data best reflects the characteristics and size of the object, thus ensuring the accuracy of the detection results.

[0185] For example, determining whether a feature point belongs to a vehicle's preset planar range may include the following steps: First, the vehicle can load a high-precision map and obtain target road information (including lane line information and curb information) within the vehicle's preset planar range from the high-precision map. Then, each feature point in the stitched point cloud data is compared with road feature points in the target road information to determine whether each feature point in the stitched point cloud data is outside the preset planar range. If a feature point in the stitched point cloud data is outside the preset planar range, then that feature point is determined to be an irrelevant feature point.

[0186] Optionally, in the target detection method provided in this embodiment of the invention, after step 203 above, if it is determined that the first object and the second object are not the same target object, it indicates that there may be a missed detection, that is, it is necessary to confirm whether the second object exists.

[0187] For example, confirming the existence of a second object may include the following steps: First, acquire second, third, and fourth point cloud data. These data are obtained based on raw point cloud data acquired by the radar within three consecutive acquisition cycles following the current acquisition cycle. Then, perform clustering processing on the second, third, and fourth point cloud data respectively to obtain a fourth bounding box representing the location of the fourth object and its category, a fifth bounding box representing the location of the fifth object and its category, and a sixth bounding box representing the location of the sixth object and its category. Finally, determine whether the second object exists based on the categories of the second, fourth, fifth, and sixth objects, and the intersection-union ratio (IUU) of each pair of bounding boxes within the second, fourth, fifth, and sixth bounding boxes.

[0188] If the first object and the second object are not the same target object, it indicates that there may be missed detections in the results obtained by the target detection model. In this case, the first object obtained by the target detection model is retained, and target tracking can be performed directly on this object. For the second object obtained through clustering, its existence needs to be confirmed again using the second, third, and fourth point cloud data collected by the radar in the three consecutive acquisition cycles after the current acquisition cycle. Specifically, clustering is performed on the second, third, and fourth point cloud data respectively to obtain three processing results. If the category in the three processing results is the same as the category of the second object, and the intersection-union ratio (IU) of any two second bounding boxes in the three processing results and the second bounding boxes of the second object is greater than the second threshold, it indicates that the second object exists, i.e., there is a missed detection. Then, target tracking can also be performed on the second object. If any category in the three processing results is different from the category of the second object, or the IU of any two second bounding boxes in the three processing results and the second bounding boxes of the second object is less than the second threshold, it indicates that the second object does not exist, i.e., there is no missed detection.

[0189] The foregoing primarily describes the solutions provided by the embodiments of the present invention from the perspective of the device. It is understood that, in order to achieve the above functions, the device includes corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the algorithmic steps of the various examples described in the embodiments disclosed herein, the present invention can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0190] Figure 6 The second schematic diagram of the target detection device 600 involved in the above embodiments is shown, as follows: Figure 6 As shown, the target detection device 600 may include: a first determining unit 601, a second determining unit 602 and a third determining unit 603.

[0191] The first determining unit 601 is used to detect a first object included in the first point cloud data using a target detection model, obtaining a first bounding box representing the location of the first object, the category of the first object, and the first target point cloud data included in the first bounding box; the first point cloud data is obtained based on the raw point cloud data collected by the radar on the vehicle in the current acquisition cycle. The second determining unit 602 is used to perform clustering processing on the first point cloud data, obtaining a second bounding box representing the location of a second object, the category of the second object, and the second target point cloud data included in the second bounding box. The third determining unit 603 is used to determine whether the first object and the second object are the same target object based on the categories of the first object and the second object, and the intersection-union ratio of the first bounding box and the second bounding box. The third determining unit 603 is further used to determine, if the first object and the second object are determined to be the same target object, a third target point cloud data for identifying the target object based on the size relationship between the first bounding box and the second target point cloud data. The third determining unit 603 is also used to identify the target object based on the third target point cloud data.

[0192] Optionally, the second determining unit 602 is specifically used for:

[0193] First feature points belonging to the target plane are identified in the first point cloud data. The target plane is the plane on which the ground is located, determined based on the target road information. The first feature points are deleted from the first point cloud data to obtain multiple second feature points. The multiple second feature points are clustered to obtain the cluster corresponding to the second object and the category label of the second object. Based on the cluster, a second bounding box representing the location of the second object is determined, as well as the second target point cloud data included in the second bounding box. Based on the category label, the category of the second object is determined.

[0194] Optionally, the second determining unit 602 is specifically used for:

[0195] Obtain target road information within a preset range of the vehicle; determine the plane equation of the target plane based on the coordinates of road feature points in the target road information; divide each feature point in the first target point cloud data into multiple grids; determine the distance between each grid and the target plane based on the feature points in each grid and the plane equation; if the distance is less than a first threshold, then each feature point in the grid is determined as the first feature point.

[0196] Optionally, the second determining unit 602 is specifically used for:

[0197] Based on the coordinates of each feature point in each grid, determine the centroid coordinates of each grid; based on each centroid coordinate and the plane equation, determine the distance between each grid and the target plane.

[0198] Optionally, the third determining unit 603 is specifically used for:

[0199] If the first object and the second object are of the same category, and the intersection-union ratio of the first bounding box and the second bounding box is greater than the second threshold, then the first object and the second object are determined to be the same target object; if the first object and the second object are of different categories, or the intersection-union ratio of the first bounding box and the second bounding box is less than the second threshold, then the first object and the second object are determined to be different target objects.

[0200] Optionally, the third determining unit 603 is specifically used for:

[0201] Determine the size relationship between the first bounding box and the second bounding box;

[0202] If the size of the second bounding box is larger than the size of the first bounding box, then the size of the second bounding box is reduced to the size of the first bounding box to obtain the third bounding box; delete the point cloud data outside the third bounding box in the second target point cloud data to obtain the fourth target point cloud data; merge the first target point cloud data and the fourth target point cloud data to obtain the third target point cloud data.

[0203] Optionally, the third determining unit 603 is also used for:

[0204] If the size of the second bounding box is smaller than the size of the first bounding box, then feature points in the second target point cloud data that do not belong to the first target point cloud data are identified as noise points; the noise points are deleted from the second target point cloud data to obtain the third target point cloud data.

[0205] Figure 7 The third schematic diagram of the target detection device 600 involved in the above embodiments is shown, as follows: Figure 7 As shown, the target detection device 600 may further include: an acquisition unit 701 and a fourth determination unit 702.

[0206] The acquisition unit 701 is used to acquire the original point cloud data of the main radar installed on the roof of the vehicle, and the original point cloud data of multiple secondary radars installed on the vehicle body. The fourth determination unit 702 is used to perform motion compensation on the original point cloud data of the main radar and the original point cloud data of the secondary radars respectively; to stitch the motion-compensated original point cloud data of the main radar and the original point cloud data of each secondary radar to obtain stitched point cloud data; and to delete irrelevant feature points located outside the preset range of the vehicle in the stitched point cloud data to obtain the first point cloud data.

[0207] Optionally, the fourth determining unit 702 is also used to determine a first transformation matrix of the vehicle relative to the target radar from the vehicle coordinate system to the map coordinate system, wherein the target radar is the main radar and each of the multiple secondary radars.

[0208] Optionally, the fourth determining unit 702 is specifically used for:

[0209] The system acquires the second transformation matrix from the map coordinate system to the main radar coordinate system, and the third transformation matrix from the main radar coordinate system to the vehicle coordinate system. Based on the first, second, and third transformation matrices, motion compensation is performed on the original point cloud data of the main radar. The system also acquires the fourth transformation matrix from the map coordinate system to the corresponding secondary radar coordinate system, the fifth transformation matrix from the secondary radar coordinate system to the vehicle coordinate system, and the sixth transformation matrix from the secondary radar coordinate system to the main radar coordinate system. Based on the first, fourth, fifth, and sixth transformation matrices, motion compensation is performed on the original point cloud data of the secondary radar.

[0210] Optionally, the aforementioned preset range is a three-dimensional spatial range. The fourth determining unit 702 is specifically used for:

[0211] Feature points in the stitched point cloud data that are outside the preset plane range of the vehicle and whose axis coordinates are outside the preset coordinate range are identified as irrelevant feature points; and irrelevant feature points are deleted from the stitched point cloud data to obtain the first point cloud data.

[0212] Optionally, the third determining unit 603 is further configured to, if it is determined that the first object and the second object are not the same target object, acquire second point cloud data, third point cloud data, and fourth point cloud data, the second point cloud data, third point cloud data, and fourth point cloud data being obtained based on the original point cloud data acquired by the radar in three consecutive acquisition cycles after the current acquisition cycle; perform clustering processing on the second point cloud data, third point cloud data, and fourth point cloud data respectively to obtain a fourth bounding box representing the location of the fourth object and the category of the fourth object, a fifth bounding box representing the location of the fifth object and the category of the fifth object, and a sixth bounding box representing the location of the sixth object and the category of the sixth object; and determine whether the second object exists based on the categories of the second object, fourth object, fifth object, and sixth object, and the intersection-union ratio of every two bounding boxes in the second bounding box, fourth bounding box, fifth bounding box, and sixth bounding box.

[0213] Of course, the target detection device 600 provided in this embodiment of the invention includes, but is not limited to, the modules described above.

[0214] In actual implementation, the first determining unit 601, the second determining unit 602, the third determining unit 603, the acquiring unit 701, and the fourth determining unit 702 can be determined by... Figure 1 The processor 11 shown calls the program code in memory 12 to implement this. For the specific execution process, please refer to... Figures 2 to 5 The description of the target detection method shown is omitted here.

[0215] Another embodiment of the present invention provides a computer-readable storage medium storing computer instructions that, when executed on a target detection device, cause the target detection device to perform each step of the method flow shown in the above method embodiment.

[0216] Another embodiment of the present invention provides a chip system applied to a target detection device. The chip system includes one or more interface circuits and one or more processors 11. The interface circuits and processors 11 are interconnected via lines. The interface circuits are used to receive signals from a memory 12 of the target detection device and send the signals to the processors 11, the signals including computer instructions stored in the memory 12. When the processors 11 execute the computer instructions, the target detection device performs each step of the method flow shown in the above method embodiment.

[0217] In another embodiment of the present invention, a computer program product is also provided, the computer program product including instructions that, when executed on a target detection device, cause the target detection device to perform each step of the method flow shown in the above method embodiment.

[0218] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software programs, implementation can be, in whole or in part, in the form of a computer program product. This computer program product includes one or more computer instructions. When these computer instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device containing one or more servers, data centers, etc., that can be integrated with the medium. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state disks, SSDs).

[0219] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions within the technical scope disclosed in the present invention should be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A target detection method, characterized in that, include: A target detection model is used to detect the first object included in the first point cloud data, resulting in a first bounding box representing the location of the first object, the category of the first object, and the first target point cloud data included in the first bounding box; the first point cloud data is obtained based on the raw point cloud data collected by the radar on the vehicle in the current acquisition cycle; Clustering is performed on the first point cloud data to obtain a second bounding box representing the location of the second object, the category of the second object, and the second target point cloud data included in the second bounding box; Based on the categories of the first object and the second object, and the intersection-union ratio of the first bounding box and the second bounding box, determine whether the first object and the second object are the same target object; If it is determined that the first object and the second object are the same target object, then based on the size relationship between the first bounding box and the second bounding box, a third target point cloud data for identifying the target object is determined from the first target point cloud data and the second target point cloud data; The target object is identified based on the third target point cloud data; The step of determining third target point cloud data for identifying the target object from the first target point cloud data and the second target point cloud data based on the size relationship between the first bounding box and the second bounding box includes: Determine the size relationship between the first bounding box and the second bounding box; If the size of the second bounding box is larger than the size of the first bounding box, then the size of the second bounding box is reduced to the size of the first bounding box to obtain the third bounding box; Delete the point cloud data located outside the third bounding box in the second target point cloud data to obtain the fourth target point cloud data; The first target point cloud data and the fourth target point cloud data are merged to obtain the third target point cloud data.

2. The target detection method according to claim 1, characterized in that, The step of clustering the first point cloud data to obtain a second bounding box representing the location of the second object, the category of the second object, and the second target point cloud data included in the second bounding box includes: Identify a first feature point in the first point cloud data that belongs to the target plane, wherein the target plane is the plane on which the ground is located, determined based on the target road information; The first feature point is deleted from the first point cloud data to obtain multiple second feature points; Clustering is performed on the multiple second feature points to obtain the cluster corresponding to the second object and the category label of the second object; Based on the cluster, a second bounding box representing the location of the second object is determined, as well as second target point cloud data included in the second bounding box, and the category of the second object is determined based on the category label.

3. The target detection method according to claim 2, characterized in that, Determining the first feature point in the first point cloud data that belongs to the target plane includes: Obtain target road information within a preset range for the vehicle; The plane equation of the target plane is determined based on the coordinates of the road feature points in the target road information; Divide each feature point in the first target point cloud data into multiple grids; Based on the feature points in each grid and the plane equation, determine the distance between each grid and the target plane; If the distance is less than a first threshold, then each feature point in the grid is identified as the first feature point.

4. The target detection method according to claim 3, characterized in that, Determining the distance between each grid and the target plane based on the feature points in each grid and the plane equation includes: The centroid coordinates of each grid are determined based on the coordinates of each feature point in each grid. The distance between each of the stated centroid coordinates and the stated plane equation is determined.

5. The target detection method according to any one of claims 1 to 4, characterized in that, Determining whether the first object and the second object are the same target object based on the categories of the first object and the second object, and the intersection-union ratio of the first bounding box and the second bounding box, includes: If the first object and the second object are of the same category, and the intersection-union ratio of the first bounding box and the second bounding box is greater than the second threshold, then the first object and the second object are determined to be the same target object. If the first object and the second object are of different categories, or if the intersection-union ratio of the first bounding box and the second bounding box is less than the second threshold, then the first object and the second object are determined to be different target objects.

6. The target detection method according to claim 1, characterized in that, After determining the size relationship between the first bounding box and the second bounding box, the target detection method further includes: If the size of the second bounding box is smaller than the size of the first bounding box, then feature points in the second target point cloud data that do not belong to the first target point cloud data are identified as noise points; The noise is removed from the second target point cloud data to obtain the third target point cloud data.

7. The target detection method according to any one of claims 1 to 4, characterized in that, Before using an object detection model to detect the first object included in the first point cloud data, the object detection method further includes: Acquire raw point cloud data from the main radar mounted on the roof of the vehicle, as well as raw point cloud data from multiple secondary radars mounted on the vehicle body. Motion compensation is performed on the raw point cloud data of the main radar and the raw point cloud data of the secondary radar, respectively. The motion-compensated original point cloud data of the main radar and the original point cloud data of each of the secondary radars are stitched together to obtain stitched point cloud data. Remove irrelevant feature points located outside the preset range of the vehicle from the stitched point cloud data to obtain the first point cloud data.

8. The target detection method according to claim 7, characterized in that, Before performing motion compensation on the raw point cloud data of the main radar and the raw point cloud data of the secondary radar, the target detection method further includes: Determine a first transformation matrix of the vehicle relative to the target radar from the vehicle coordinate system to the map coordinate system, wherein the target radar is the main radar and each of the plurality of secondary radars.

9. The target detection method according to claim 8, characterized in that, The motion compensation of the raw point cloud data of the main radar includes: Obtain the second transformation matrix of the main radar from the map coordinate system to the main radar coordinate system, and the third transformation matrix of the main radar from the main radar coordinate system to the vehicle coordinate system; Motion compensation is performed on the original point cloud data of the main radar based on the first transformation matrix, the second transformation matrix and the third transformation matrix of the main radar. The motion compensation of the raw point cloud data of the secondary radar includes: Obtain the fourth transformation matrix of the secondary radar from the map coordinate system to the corresponding secondary radar coordinate system, the fifth transformation matrix of the secondary radar from the secondary radar coordinate system to the vehicle coordinate system, and the sixth transformation matrix of the secondary radar from the secondary radar coordinate system to the main radar coordinate system; Motion compensation is performed on the original point cloud data of the secondary radar based on the first transformation matrix, the fourth transformation matrix, the fifth transformation matrix, and the sixth transformation matrix.

10. The target detection method according to claim 8 or 9, characterized in that, The preset range is a three-dimensional spatial range. Deleting irrelevant feature points located outside the preset range of the vehicle from the stitched point cloud data to obtain the first point cloud data includes: Feature points in the stitched point cloud data that are outside the preset plane range of the vehicle, and feature points whose axis coordinates are outside the preset coordinate range, are determined as irrelevant feature points. The irrelevant feature points are deleted from the stitched point cloud data to obtain the first point cloud data.

11. The target detection method according to claim 8 or 9, characterized in that, After determining whether the first object and the second object are the same target object, the target detection method further includes: If it is determined that the first object and the second object are not the same target object, then the second point cloud data, the third point cloud data, and the fourth point cloud data are obtained. The second point cloud data, the third point cloud data, and the fourth point cloud data are respectively obtained based on the original point cloud data collected by the radar in three consecutive collection cycles after the current collection cycle. Clustering is performed on the second point cloud data, the third point cloud data, and the fourth point cloud data respectively to obtain a fourth bounding box representing the location of the fourth object and the category of the fourth object, a fifth bounding box representing the location of the fifth object and the category of the fifth object, and a sixth bounding box representing the location of the sixth object and the category of the sixth object; The existence of the second object is determined based on the categories of the second object, the fourth object, the fifth object, and the sixth object, as well as the intersection-union ratio of every two bounding boxes in the second bounding box, the fourth bounding box, the fifth bounding box, and the sixth bounding box.

12. A target detection device, characterized in that, include: The first determining unit is used to detect a first object included in the first point cloud data using a target detection model, and obtain a first bounding box representing the location of the first object, the category of the first object, and the first target point cloud data included in the first bounding box; the first point cloud data is obtained based on the raw point cloud data collected by the radar on the vehicle in the current acquisition cycle; The second determining unit is used to perform clustering processing on the first point cloud data to obtain a second bounding box representing the location of the second object, the category of the second object, and the second target point cloud data included in the second bounding box; The third determining unit determines whether the first object and the second object are the same target object based on the categories of the first object and the second object, and the intersection-union ratio of the first bounding box and the second bounding box. The third determining unit is further configured to, if it is determined that the first object and the second object are the same target object, determine third target point cloud data for identifying the target object from the first target point cloud data and the second target point cloud data when the size of the second bounding box is larger than the size of the first bounding box; The third determining unit is further configured to identify the target object based on the third target point cloud data; When the size of the second bounding box is larger than the size of the first bounding box, and when determining the third target point cloud data for identifying the target object from the first target point cloud data and the second target point cloud data, the third determining unit is specifically configured to perform: Determine the size relationship between the first bounding box and the second bounding box; If the size of the second bounding box is larger than the size of the first bounding box, then the size of the second bounding box is reduced to the size of the first bounding box to obtain the third bounding box; Delete the point cloud data located outside the third bounding box in the second target point cloud data to obtain the fourth target point cloud data; The first target point cloud data and the fourth target point cloud data are merged to obtain the third target point cloud data.

13. A target detection device, characterized in that, The target detection device includes a processor and a memory; the memory is used to store computer program code, the computer program code including computer instructions; when the processor executes the computer instructions, the target detection device performs the target detection method as described in any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that, The device includes computer instructions that, when executed on the target detection device, cause the target detection device to perform the target detection method according to any one of claims 1 to 11.

Images