Obstacle detection methods, lidar systems, and computer-readable storage media

Abstract

This application discloses an obstacle detection method, a lidar system, and a computer-readable storage medium. The method includes: acquiring point clouds collected sequentially by a first radar and a second radar over N acquisition cycles, where N is an integer greater than 1; obtaining a main grid map for the Nth acquisition cycle based on the point cloud collected by the first radar in the Nth acquisition cycle, the point cloud collected by the second radar in the Nth acquisition cycle, and the main grid map of the (N-1)th acquisition cycle; and performing cluster analysis on the main grid map of the Nth acquisition cycle to obtain obstacle information within the field of view of the second radar. By fusing the perception results of the first radar as prior knowledge with the point cloud of the second radar, and combining it with historical obstacle information containing temporal sequence, the obstacle detection accuracy of the second radar's field of view can be improved while saving computational power.

Description

Technical Field

[0001] This application relates to the field of lidar technology, and in particular to an obstacle detection method, a lidar system, and a computer-readable storage medium. Background Technology

[0002] With the development of science and technology, lidar is widely used in intelligent equipment such as autonomous driving, intelligent robot navigation, and drones. LiDAR is a sensing device that uses laser beams to detect the position, velocity, and other characteristics of a target object. Its working principle is to first emit a detection laser beam towards the target object, then compare the received reflected signal from the target object with the emitted signal, process it, and obtain relevant information about the target object.

[0003] In autonomous or assisted driving modes, a vehicle's perception of its surroundings is crucial, typically employing a LiDAR system with multiple LiDAR sensors to provide full-range perception capabilities from near to far. However, current traditional obstacle perception algorithms and LiDAR systems struggle to balance detection accuracy with computational costs. Summary of the Invention

[0004] In view of this, embodiments of this application provide an obstacle detection method, a lidar system, and a computer-readable storage medium, which improves the accuracy of obstacle perception while effectively saving computing power.

[0005] In a first aspect, embodiments of this application provide an obstacle detection method applied to a lidar system, the lidar system including a first radar and a second radar. The method includes: acquiring point clouds collected by the first radar and the second radar sequentially over N acquisition cycles, where N is an integer greater than 1; obtaining a main grid map for the Nth acquisition cycle based on the point cloud collected by the first radar in the Nth acquisition cycle, the point cloud collected by the second radar in the Nth acquisition cycle, and the main grid map for the (N-1)th acquisition cycle, wherein the main grid map for the (N-1)th acquisition cycle is determined based on the point clouds collected by the first radar and the second radar in the previous N-1 acquisition cycles; and performing cluster analysis on the main grid map for the Nth acquisition cycle to obtain obstacle information within the field of view of the second radar.

[0006] In the above technical solution, the point clouds acquired by the first and second radars in the first N-1 acquisition cycles represent the historical obstacle information sensed by the first and second radars. Since the field of view areas of the first and second radars overlap temporally, the main grid map of the N-1 acquisition cycle is combined with the point clouds acquired by the first and second radars in the Nth acquisition cycle to determine the main grid map of the Nth acquisition cycle. This allows the historical obstacle information sensed by the first radar to be used as prior knowledge, enabling the main grid map of the Nth acquisition cycle to more accurately reflect the obstacle information within the field of view of the second radar, thus facilitating high-precision obstacle detection within the field of view of the second radar. On the one hand, using the historical obstacle information sensed by the first radar as prior knowledge can improve the perception accuracy of obstacles within the field of view of the second radar and effectively save computing power. On the other hand, by generating the main grid map for the Nth acquisition cycle, the three-dimensional point cloud is reduced to a two-dimensional grid map, and obstacles are identified in two-dimensional space, which can effectively reduce the time and space complexity of the target detection algorithm.

[0007] In some embodiments, the aforementioned process of obtaining the main grid map for the Nth acquisition cycle based on the point cloud acquired by the first radar in the Nth acquisition cycle, the point cloud acquired by the second radar in the Nth acquisition cycle, and the main grid map for the (N-1)th acquisition cycle includes: obtaining a first radar observation grid layer based on the point cloud acquired by the first radar in the Nth acquisition cycle; obtaining a second radar observation grid layer based on the point cloud acquired by the second radar in the Nth acquisition cycle; obtaining a main grid map observation layer for the Nth acquisition cycle based on the first radar observation grid layer and the second radar observation grid layer; and obtaining the main grid map for the Nth acquisition cycle based on the main grid map observation layer for the Nth acquisition cycle and the main grid map for the (N-1)th acquisition cycle.

[0008] In the above technical solution, the first radar observation grid layer and the second radar observation grid layer are superimposed to obtain the main grid map observation layer. The main grid map for the N-1th acquisition cycle is updated through the main grid map observation layer to obtain the main grid map for the Nth acquisition cycle, thus giving the main grid map a temporal sequence. The main grid map layer is updated in real time using the point cloud acquired by the second radar in the Nth acquisition cycle and the point cloud acquired by the second radar in the Nth acquisition cycle, which can reflect the current obstacle information within the field of view of the second radar. The introduced temporal information is beneficial for the accurate identification of obstacles within the field of view of the second radar.

[0009] In some embodiments, obtaining the first radar observation grid layer based on the point cloud acquired by the first radar during the Nth acquisition cycle includes: rasterizing the point cloud acquired by the first radar during the Nth acquisition cycle to obtain a first grid, wherein the first grid is marked with a grid height and a height difference, the grid height being the median height of the point cloud points in the grid, and the height difference being the absolute value of the difference between the maximum and minimum heights of the point cloud points in the grid; marking the obstacle attributes of the first grid based on the obstacle information of the point cloud acquired by the first radar during the Nth acquisition cycle, wherein the obstacle attributes include obstacle grids or non-obstacle grids; determining and marking the region to which the first grid belongs based on the grid height, height difference, and obstacle attributes of the first grid to obtain the first radar observation grid layer, wherein the region includes a region of interest or a region of non-interest.

[0010] In the above technical solution, the grid layer corresponding to the point cloud acquired by the first radar in the Nth acquisition cycle is divided into the area of ​​interest and the area of ​​non-interest by using the grid height, height difference and obstacle attributes. This distinguishes between obstacles in the field of view of the first radar and fixed objects on the roadside such as green belts or steps, which is conducive to accurately identifying obstacles in the field of view of the second radar and reducing interference.

[0011] In some embodiments, obtaining a second radar observation grid layer based on the point cloud acquired by the second radar during the Nth acquisition period includes: rasterizing the point cloud acquired by the second radar during the Nth acquisition period to obtain a second grid, wherein the second grid is marked with a grid height and a height difference; determining and marking the ground attributes of the second grid based on the grid height and height difference of the second grid to obtain the second radar observation grid layer, wherein the ground attributes include ground grids or non-ground grids.

[0012] In the above technical solution, combining grid height and height difference to distinguish between ground and non-ground areas within the field of view of the second radar helps to avoid interference from ground areas in obstacle perception.

[0013] In some embodiments, determining and marking the region to which the first grid belongs based on the grid height, height difference, and obstacle attributes of the first grid includes: determining and marking the region to which the first grid belongs that meets a first condition as a non-interested sensing region, and determining and marking the region to which the first grid belongs that does not meet the first condition as an interested sensing region; wherein, the first condition is that the grid height is greater than or equal to a first threshold, the height difference is greater than or equal to a second threshold, and the obstacle attribute is a non-obstacle grid, and the first threshold and the second threshold are determined based on the point cloud acquired by the first radar in the first N-1 acquisition cycles.

[0014] In the above technical solution, the area to which the first grid belongs is segmented by using a first threshold and a second threshold, thereby focusing on the perception area and improving the targeting and accuracy of obstacle detection.

[0015] In some embodiments, the aforementioned determination and labeling of the ground attributes of the second grid based on the grid height and height difference of the second grid includes: determining and labeling the second grid that meets the second condition as a non-ground grid, and determining and labeling the second grid that does not meet the second condition as a ground grid; wherein the second condition is that the grid height is greater than or equal to a third threshold, or the height difference is greater than or equal to a fourth threshold, and the third threshold and the fourth threshold are determined based on the point cloud acquired by the second radar in the first N-1 acquisition cycles.

[0016] In the above technical solution, the ground attribute to which the second grid belongs is determined by the third threshold and the fourth threshold, thereby accurately dividing the ground grid and non-ground grid and improving the targeting and accuracy of obstacle detection.

[0017] In some embodiments, obtaining the main grid map observation layer based on the first radar observation grid layer and the second radar observation grid layer includes: superimposing the first radar observation grid layer and the second radar observation grid layer to obtain a fused grid map layer; determining and marking the obstacle attributes of the fused grid map layer based on the region to which the first radar observation grid layer belongs and the ground attributes of the second radar observation grid layer to obtain the main grid map observation layer.

[0018] In the above technical solution, the obstacle attributes of the fused grid map layer are determined by the dual determination of the regional attributes of the first radar observation grid layer and the ground attributes of the second radar observation grid layer, which helps to improve the accuracy and reliability of obstacle detection.

[0019] Secondly, embodiments of this application provide a lidar system, including:

[0020] Memory, used to store executable program code;

[0021] The processor is used to call and run executable program code from memory, enabling the lidar system to perform the obstacle detection method described in the first aspect above.

[0022] Thirdly, some embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed, implements the obstacle detection method as described in the first aspect. Attached Figure Description

[0023] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0024] Figure 1 This is a schematic diagram of the application environment of the obstacle detection method in some embodiments of this application;

[0025] Figure 2 These are schematic diagrams of the lidar system in some embodiments of this application;

[0026] Figure 3 This is a flowchart illustrating the obstacle detection method in some embodiments of this application;

[0027] Figure 4 This is a schematic diagram of the field of view of the first and second radars in some embodiments of this application;

[0028] Figure 5 This is a flowchart illustrating the obstacle detection method in some other embodiments of this application.

[0029] Figure Labels

[0030] 300. Driving environment; 301. Obstacles;

[0031] 200. Vehicles;

[0032] 100. LiDAR system;

[0033] 10. First radar, 20. Second radar, 40. Processor, 50. Memory, 60. Bus. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and embodiments. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0035] Most obstacle detection schemes known to the inventors of this application employ traditional point cloud clustering-based target detection algorithms or deep learning-based perception algorithms to process 3D point clouds for obstacle identification. However, traditional point cloud clustering-based target detection algorithms have low accuracy and robustness, and cannot guarantee good timeliness; deep learning-based perception algorithms have high training costs, require significant computing power during operation, and are more sensitive to data. For autonomous vehicles, the allocation of sensors and computing power is biased, with forward and backward perception tasks generally receiving a larger proportion of computing power and cost, while lateral blind spot detection tasks often receive less computing power support. Consequently, with limited hardware configuration and processor computing power, the timeliness and accuracy of lateral obstacle detection are difficult to meet the expected requirements.

[0036] To address the aforementioned technical problems, embodiments of this application provide an obstacle detection method, a lidar system, and a computer-readable storage medium, which improves the accuracy of obstacle perception while effectively saving computing power.

[0037] The following example, using a lidar system applied to an autonomous vehicle, illustrates the application scenarios of the lidar system in this application. Figure 1 As shown, the lidar system 100 is installed on the vehicle 200. The vehicle 200 uses the lidar system 100 to detect obstacles 301 in the driving environment 300, enabling it to plan its driving path based on the detected information to avoid collisions with obstacles 301. It is understood that obstacles 301 can be other vehicles, pedestrians, bicycles, or electric motorcycles, etc., and the vehicle 200 can be a passenger car, a freight truck, a cleaning vehicle, or a detection vehicle, etc. Since the field of view of a single lidar is limited, the lidar system 100 includes multiple lidars, which can be configured as needed to perform multi-directional or all-directional detection of obstacles around the vehicle 200.

[0038] The obstacle detection method in this application embodiment is executed by the processor of the LiDAR system. Before introducing the obstacle detection method in this application embodiment, the LiDAR system provided in this application embodiment will be described in detail first.

[0039] Please see Figure 2 The lidar system 100 includes a first radar 10, a second radar 20, a processor 40, and a memory 50. The first radar 10, the second radar 20, the processor 40, and the memory 50 are communicatively connected via a bus 60.

[0040] The memory 50 is used to store executable program code. As a non-volatile computer-readable storage medium, the memory 50 can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the obstacle detection method in this embodiment. In some embodiments, the memory 50 includes a program storage area and a data storage area, wherein the program storage area is used to store applications required by the operating system or at least one function, and the data storage area is used to store data created based on the use of the terminal. In some embodiments, the memory 50 may include high-speed random access memory or non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 50 includes a remote memory remotely disposed relative to the processor 40, which is used to connect to the terminal via a network. The aforementioned network includes, but is not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0041] Processor 40 is used to call and run executable program code from memory 50, causing lidar system 100 to perform the obstacle detection method in any embodiment of this application. In some embodiments, processor 40 includes one or more of a central processing unit (CPU), digital signal processor (DSP), application-specific integrated circuit (ASIC), controller, or field-programmable gate array (FPGA) device. Processor 40 may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors incorporating a DSP core, or any other such configuration.

[0042] In some embodiments, the lidar system 100 further includes other supporting hardware and software. The hardware may include a power supply, antenna, temperature control system, or a Global Positioning System (GPS) / Inertial Measurement Unit (IMU) integrated navigation system. The software may include radar driver software, system control software, data processing software, or Simultaneous Localization and Mapping (SLAM) software. Specifically, the radar driver software drives the radar hardware to acquire data; the system control software controls the setting of radar parameters such as scanning mode, frequency, and field of view; the data processing software performs signal filtering, time synchronization, or coordinate transformation; and the SLAM software is used for localization and map building.

[0043] Understandable, Figure 2The example only uses one first radar 10 and one second radar 20. The number of first radars 10 and the number of second radars 20 can be one or more, and the number of radars in the lidar system and the field of view of each radar can be set according to actual conditions. The first radar is used to detect obstacles in front of or behind the vehicle, and the second radar is used to detect obstacles to the side of the vehicle. In some embodiments, the detection range and accuracy of the first radar are higher than those of the second radar.

[0044] In some embodiments, the lidar system 100 includes two first radars and two second radars, wherein the two first radars have the same horizontal field of view and are oriented in opposite directions. In some embodiments, the lidar system 100 includes one first radar and two second radars, wherein the first radar is used for forward mid-to-long-range detection, and the two second radars are used for left and right side-to-side short-range detection, respectively.

[0045] In one embodiment, the first radar has a horizontal field of view (HFOV) of 120° and a vertical field of view (VFOV) of 25°. The second radar has a horizontal field of view (HFOV) of 180° and a vertical field of view (VFOV) of 90°.

[0046] The obstacle detection method provided in this application will be described in detail below with reference to the exemplary application and implementation of the lidar system 100 provided in the embodiments of this application. Please refer to [link to relevant documentation]. Figure 3 The obstacle detection method S100 includes, but is not limited to, the following steps:

[0047] S10: Obtain the point cloud acquired by the first radar and the second radar in N acquisition cycles in a sequential order, where N is an integer greater than 1.

[0048] The aforementioned acquisition cycle can be preset according to actual needs. In one possible implementation, the time required for the first or second radar to acquire one frame of point cloud is defined as an acquisition cycle. Correspondingly, the point cloud acquired by the first radar within one acquisition cycle can be understood as one frame of point cloud acquired by the first radar, and the point cloud acquired by the second radar within one acquisition cycle can be understood as one frame of point cloud acquired by the second radar. The system control software of the lidar system can control the first and second radars to scan at the same frame rate, generating a corresponding point cloud frame with each scan, and synchronizing the point clouds acquired by the first and second radars in time. For example, the frame rate can be 600 frames / second. The point clouds acquired by the first and second radars sequentially over N acquisition cycles include multiple frames of point cloud acquired by the first radar in chronological order, and multiple frames of point cloud acquired by the second radar in chronological order. The point cloud acquired by the first radar in the Nth acquisition cycle refers to the latest frame of point cloud generated by the first radar scan at the current moment, and the point cloud acquired by the second radar in the Nth acquisition cycle refers to the latest frame of point cloud generated by the second radar scan at the current moment. When the lidar system 100 is running, the processor 40 performs calculations on the point clouds acquired by the first and second radars in N acquisition cycles at a certain frequency. The current time can refer to any time when the processor 40 performs calculations.

[0049] S20: Based on the point cloud acquired by the first radar in the Nth acquisition cycle, the point cloud acquired by the second radar in the Nth acquisition cycle, and the main grid map in the N-1th acquisition cycle, the main grid map of the Nth acquisition cycle is obtained. The main grid map of the N-1th acquisition cycle is determined based on the point cloud acquired by the first radar and the second radar in the previous N-1 acquisition cycles.

[0050] The point cloud acquired by the first radar in the first N-1 acquisition cycles includes multiple frames of point cloud whose temporal sequence precedes the point cloud acquired in the Nth acquisition cycle; the point cloud acquired by the second radar in the first N-1 acquisition cycles also includes multiple frames of point cloud whose temporal sequence precedes the point cloud acquired in the Nth acquisition cycle. It can be understood that the main grid map is continuously updated with each acquisition cycle; specifically, the update frequency of the main grid map is the same as the frame rate of the point cloud. The main grid map in the N-1th acquisition cycle refers to the main grid map corresponding to the point cloud acquired by either the first radar or the second radar in the N-1th acquisition cycle. The main grid map in the Nth acquisition cycle refers to the main grid map corresponding to the point cloud acquired by either the first radar or the second radar in the Nth acquisition cycle. Since the point cloud collected by the first radar in the first N-1 acquisition cycles can reflect the historical obstacle information within the field of view of the first radar, and the point cloud collected by the second radar in the first N-1 acquisition cycles can reflect the historical obstacle information within the field of view of the second radar, the main grid map in the N-1 acquisition cycle can reflect the historical obstacle information within the field of view of both the first radar and the second radar.

[0051] It is understandable that, on the one hand, in some embodiments, the point cloud acquired by the first radar in the Nth acquisition cycle and the point cloud acquired by the second radar in the Nth acquisition cycle overlap in the field of view. The point cloud acquired by the first radar in the Nth acquisition cycle can reflect obstacle information in part of the field of view of the second radar. Therefore, the point cloud acquired by the first radar in the Nth acquisition cycle can be used to identify obstacles in the field of view of the second radar. On the other hand, since the lidar system is mounted on a mobile device, as the mobile device moves, the field of view of the first radar and the field of view of the second radar overlap in time. This leads to the application of the lidar system to... Figure 1 Taking the illustrated application environment as an example, as vehicle 200 moves forward, the field of view of the first radar will continuously fall into the field of view of the second radar. That is, an obstacle to the left or right front of vehicle 200 that is currently within the scanning range of the first radar will enter the scanning range of the second radar at the next moment or some future moment. Therefore, the current obstacle information within the field of view of the second radar can be determined by referring to the historical obstacle information within the field of view of the first radar. The main grid map in the (N-1)th acquisition cycle can effectively compensate for areas in the point cloud acquired by the second radar in the Nth acquisition cycle that do not intersect with the point cloud acquired by the first radar in the Nth acquisition cycle, i.e., effectively compensate for areas in the field of view of the second radar that do not intersect with the field of view of the first radar.

[0052] It can be seen that, as Figure 4As shown, for a first region where the point clouds acquired by the first radar in the Nth acquisition cycle and the point clouds acquired by the second radar in the Nth acquisition cycle intersect in the field of view, obstacle information within this first region can be determined by referring to the point clouds acquired by the first radar in the Nth acquisition cycle and the point clouds acquired by the second radar in the Nth acquisition cycle. For a second region where the point clouds acquired by the first radar in the Nth acquisition cycle and the point clouds acquired by the second radar in the Nth acquisition cycle do not intersect in the field of view, obstacle information within this second region can be determined by referring to the main grid map in the (N-1)th acquisition cycle. This ensures that the main grid map in the Nth acquisition cycle accurately reflects the obstacle information within the field of view of the second radar.

[0053] S30: Perform cluster analysis on the main grid map of the Nth acquisition cycle to obtain obstacle information within the field of view of the second radar.

[0054] Since the field of view of the first radar and the field of view of the second radar overlap in time, the historical obstacle information provided by the main grid map in the (N-1)th acquisition cycle is used as prior knowledge. Combined with the point cloud collected by the first radar in the Nth acquisition cycle and the point cloud collected by the second radar in the Nth acquisition cycle, a main grid map in the Nth acquisition cycle that can reflect the obstacle information in the field of view of the second radar is obtained. This is conducive to the real-time and accurate identification of obstacles in the field of view of the second radar and the subsequent tracking of obstacles based on the time sequence information.

[0055] Cluster analysis is performed on the main grid map during the Nth acquisition cycle to obtain obstacle information within the field of view of the second radar. Cluster analysis refers to using a clustering algorithm to divide the obstacle grids in the main grid map during the Nth acquisition cycle into multiple groups based on the distance between obstacle grids. Each group represents one obstacle. The clustering algorithm can be K-means, hierarchical clustering, or meanshift, etc. It is understood that a group represents one obstacle or multiple overlapping obstacles. In some embodiments, the center coordinates of the group and its outline shape on the map can be used as the corresponding obstacle information.

[0056] In some embodiments, bounding boxes are fitted to the clustering results (the aforementioned multiple groups). Specifically, the minimum bounding rectangle of each group is generated as the bounding box. In this embodiment, the center coordinates and length and width of the bounding box can be used as the corresponding obstacle information. In some embodiments, the position of the obstacle point cloud within the bounding box can be optimized to obtain more accurate obstacle information. The position optimization can be performed using the Axis-Aligned Bounding Box (ABB) optimization algorithm.

[0057] Based on the clustering analysis results, obstacle information within the field of view of the second radar is determined. On the one hand, the historical obstacle information sensed by the first radar is used as prior knowledge, which can effectively avoid the interference of non-dynamic obstacles on the clustering analysis of dynamic obstacles. On the other hand, by generating the main grid map of the Nth acquisition cycle, the three-dimensional point cloud is reduced to a two-dimensional grid map. Obstacles are identified in two-dimensional space, which can effectively reduce the time and space complexity of the identification algorithm, thereby saving computing power.

[0058] In some embodiments, the main grid map for the Nth acquisition cycle is obtained based on the point cloud acquired by the first radar in the Nth acquisition cycle, the point cloud acquired by the second radar in the Nth acquisition cycle, and the main grid map for the (N-1)th acquisition cycle. Specifically, this includes:

[0059] S21: Based on the point cloud acquired by the first radar during the Nth acquisition cycle, obtain the first radar observation grid layer; based on the point cloud acquired by the second radar during the Nth acquisition cycle, obtain the second radar observation grid layer.

[0060] As the data acquisition cycle is continuously updated, the point cloud from the Nth acquisition cycle is typically used to guide mobile devices (e.g., vehicle 200) in obstacle avoidance or path planning. That is, the point cloud from the Nth acquisition cycle is the observation point cloud. In some embodiments, the data used for current observation and sensing can be divided into observation layers. Exemplarily, the observation layer includes the point cloud from the Nth acquisition cycle or other data (e.g., a raster map) generated based on the point cloud from the Nth acquisition cycle.

[0061] The raster map obtained by rasterizing the point cloud acquired by the first radar during the Nth acquisition cycle is the first radar observation raster layer. Here, the first radar observation raster layer refers to the raster map of the first radar that is currently used for observation and is the most up-to-date in time. Similarly, the raster map obtained by rasterizing the point cloud acquired by the second radar during the Nth acquisition cycle is the second radar observation raster layer. Here, the second radar observation raster layer refers to the raster map of the second radar that is currently used for observation and is the most up-to-date in time.

[0062] S22: Based on the first radar observation grid layer and the second radar observation grid layer, obtain the main grid map observation layer for the Nth acquisition cycle.

[0063] The first and second radar observation grid layers are merged to obtain the main grid map observation layer for the Nth acquisition cycle. This main grid map observation layer for the Nth acquisition cycle includes the regions to which each grid cell belongs, as recorded in the first radar observation grid layer, and the ground attributes of each grid cell, as recorded in the second radar observation grid layer. It can be understood that for fields of view that do not overlap between the first and second radars, the first and second radar observation grid layers are stitched together; for fields of view that overlap between the first and second radars, the regional attributes of the grid cells in the first radar observation grid layer and the ground attributes of the grid cells in the second radar observation grid layer can be combined. As the first and second radar observation grid layers are continuously updated with the point cloud acquisition cycle, the main grid map observation layer is also continuously updated with the point cloud acquisition cycle.

[0064] S23: Based on the observation layer of the main grid map in the Nth acquisition cycle and the main grid map in the (N-1)th acquisition cycle, obtain the main grid map for the Nth acquisition cycle.

[0065] The main grid map of the Nth acquisition cycle, obtained by combining the observation layer of the main grid map of the Nth acquisition cycle and the main grid map of the (N-1)th acquisition cycle, has temporal sequence and can reflect the obstacle information in the field of view of the second radar. It is also beneficial for subsequent tracking of dynamic obstacles in the field of view of the second radar.

[0066] In some embodiments, such as Figure 4 As shown, the main grid map covers an area within 10 meters in front of and behind the vehicle and 20 meters to the left and right. This area includes part of the field of view of the first radar and at least part of the field of view of the second radar. When the lidar system is started, the main grid map is a blank grid map. As the point cloud acquisition cycle is updated, the main grid map for the Nth acquisition cycle is generated based on the observation layer of the main grid map for the Nth acquisition cycle and the main grid map for the (N-1)th acquisition cycle. For example, the main grid map for the 0th acquisition cycle is the initial blank grid map, the main grid map for the 1st acquisition cycle is the observation layer of the main grid map for the 1st acquisition cycle, the main grid map for the 2nd acquisition cycle is obtained based on the observation layer of the main grid map for the 2nd acquisition cycle and the main grid map for the 1st acquisition cycle, the main grid map for the 3rd acquisition cycle is obtained based on the observation layer of the main grid map for the 3rd acquisition cycle and the main grid map for the 2nd acquisition cycle, and so on.

[0067] Specifically, in some embodiments, grid k in the main grid map observation layer of the Nth acquisition cycle and grid k' in the main grid map of the (N-1)th acquisition cycle represent the same location in the field of view of the lidar system. Then, the obstacle attributes of grid k' are updated based on the obstacle attributes of grid k and grid k'. Once the obstacle attributes of all grids in the main grid map of the (N-1)th acquisition cycle have been updated, the main grid map of the Nth acquisition cycle is obtained. In some embodiments, a static binary Bayesian filter is used to calculate the updated obstacle attributes based on the obstacle attributes of grid k and grid k'.

[0068] In some embodiments, a first radar observation grid layer is obtained based on the point cloud acquired by the first radar during the Nth acquisition cycle, including:

[0069] S211: Rasterize the point cloud acquired by the first radar in the Nth acquisition cycle to obtain the first grid.

[0070] Rasterization refers to converting discrete point cloud data into regular raster data. Specifically, rasterization includes the following process: Projecting the point cloud acquired by the first radar in the Nth acquisition cycle onto the xy plane to form a two-dimensional point cloud containing height values. Then, dividing the xy plane containing the two-dimensional point cloud into a r*r grid, obtaining the raster layer corresponding to the point cloud acquired by the first radar in the Nth acquisition cycle. Here, r is a set value that can be set according to the actual situation, for example, r is 0.1m. When converting the point cloud acquired by the first radar in the Nth acquisition cycle of the three-dimensional state into a two-dimensional point cloud, the height value of each laser point in the point cloud acquired by the first radar in the Nth acquisition cycle is retained as the height value of the two-dimensional point cloud, and the z-axis direction is perpendicular to the xy plane.

[0071] In some embodiments, the first grid is any grid in the grid layer corresponding to the point cloud acquired by the first radar. The first grid is marked with a grid height and a height difference, and the two-dimensional point cloud in the first grid includes multiple laser points. In some embodiments, the median height of the point cloud points in the grid can be used as the grid height. The grid height reflects the height of the grid in the coordinate system of the lidar system. In some embodiments, the absolute value of the difference between the maximum and minimum heights of the point cloud points in the grid can be used as the grid height difference. The height difference reflects the distance between the highest and lowest points of the object corresponding to the grid, that is, it reflects the height of the object.

[0072] S212: Based on the obstacle information of the point cloud acquired by the first radar in the Nth acquisition cycle, mark the obstacle attributes of the first grid.

[0073] The processor of the lidar system uses a deep learning-based perception algorithm to detect obstacles in each frame of point cloud acquired by the first lidar, obtaining real-time obstacle information ahead for obstacle avoidance or route planning. In some embodiments, the perception algorithm may be the PointPillars algorithm or other pre-trained neural network algorithms.

[0074] After the point cloud acquired by the first radar in the Nth acquisition cycle is processed by the perception algorithm, obstacle information is marked in the point cloud acquired by the first radar in the Nth acquisition cycle. For example, the perception algorithm identifies point cloud features to generate detection boxes, which are labeled with obstacle categories. The detection boxes are represented by positional parameters (x, y, z, w, l, h, θ), where (x, y, z) are the center coordinates of the detection box, w is the width of the detection box, l is the length of the detection box, h is the height of the detection box, and θ is the pose of the detection box. The point cloud enclosed by the detection box represents the obstacle of the corresponding category. For example, if the obstacle category of detection box 1# is vehicle, then the point cloud enclosed by detection box 1# is identified as a vehicle; if the obstacle category of detection box 2# is pedestrian, then the point cloud enclosed by detection box 2# is identified as a pedestrian; if the obstacle category of detection box 3# is cyclist, then the point cloud enclosed by detection box 3# is identified as a cyclist.

[0075] In this embodiment, point clouds enclosed by each detection frame are uniformly marked as obstacles (e.g., 1 represents an obstacle), while point clouds outside the detection frames are uniformly marked as non-obstacles (e.g., 0 represents a non-obstacle), ensuring that the point clouds acquired by the first radar in the Nth acquisition cycle contain obstacle information. Therefore, based on the obstacle information of the point clouds acquired by the first radar in the Nth acquisition cycle, the obstacle attributes of the first grid are marked. These obstacle attributes include obstacle grids or non-obstacle grids. It is understood that an obstacle grid refers to a grid occupied by an obstacle, and a non-obstacle grid refers to a grid not occupied by an obstacle.

[0076] Specifically, in some embodiments, the first grid is marked as an obstacle grid when the point cloud in the first grid is marked as an obstacle or when more than a first percentage of the point cloud in the first grid is marked as an obstacle. The first percentage can be set by those skilled in the art based on actual conditions; for example, the first percentage may be 60% or 70%. In some embodiments, the first grid is marked as a non-obstacle grid when the point cloud in the first grid is not marked as an obstacle or when more than a second percentage of the point cloud in the first grid is not marked as an obstacle. The second percentage can be set by those skilled in the art based on actual conditions; for example, the second percentage may be 60% or 50%.

[0077] S213: Based on the grid height, height difference, and obstacle attributes of the first grid, determine and mark the area to which the first grid belongs to obtain the first radar observation grid layer.

[0078] The regions are categorized into areas of interest and areas of non-interest. An area of ​​interest refers to the region within the first radar observation grid layer where obstacle detection is required. For example, an area of ​​interest typically includes obstacles, passable surfaces, or other objects in a road that impede passage. An area of ​​non-interest refers to the region within the first radar observation layer where obstacle detection is not required. For example, a non-interest area typically includes fixed objects along the roadside such as green belts or steps.

[0079] The grid height and height difference reflect the height of the object corresponding to the first grid cell, while obstacle attributes indicate whether the object is an obstacle. Combining the height and obstacle attributes, it can be determined whether the object corresponding to the first grid cell is a fixed roadside object such as a green belt or steps, thus determining the area to which the first grid cell belongs. It can be understood that for any first grid cell in the grid layer corresponding to the point cloud acquired by the first radar in the Nth acquisition cycle, its area is determined in the above manner. After the areas to which each first grid cell in the grid layer corresponding to the point cloud acquired by the first radar in the Nth acquisition cycle belongs are determined, the first radar observation grid layer is obtained.

[0080] In autonomous driving scenarios, perception algorithms are typically used to detect dynamic obstacles such as vehicles, pedestrians, or cyclists. Fixed roadside objects, such as green belts or steps, are usually considered objects that do not need to be detected. However, when detecting green belts or steps is necessary, the perception algorithm requires further training and optimization, resulting in complex algorithms and high computational costs. The method described above divides the grid layer corresponding to the point cloud acquired by the first radar in the Nth acquisition cycle into regions of interest and regions of non-interest. This reduces the interference of fixed roadside objects on obstacles in the field of view of the second radar, facilitating accurate obstacle perception in the subsequent second radar field of view. Furthermore, compared to using perception algorithms to identify objects such as green belts or steps, this method is computationally simpler and effectively saves computational resources.

[0081] In some embodiments, determining and marking the area to which the first grid belongs based on the grid height, height difference, and obstacle attributes of the first grid to obtain a first radar observation grid layer includes: determining and marking the area to which the first grid belongs that meets the first condition as a non-interested sensing area, and determining and marking the area to which the first grid belongs that does not meet the first condition as an interested sensing area.

[0082] In some embodiments, a value of 0 can be used to represent the aforementioned non-interested sensing area, and a value of 1 can be used to represent the aforementioned interested sensing area. That is, when the area to which the first grid belongs is 1, it means that it belongs to the interested sensing area; when the area to which the first grid belongs is 0, it means that it belongs to the non-interested sensing area. It can be understood that each grid in the main grid map of the Nth acquisition cycle is marked with obstacle attributes. Obstacle grids represent that the corresponding object is a target obstacle, and non-obstacle grids represent that the corresponding object is not a target obstacle. For example, in the main grid map of the Nth acquisition cycle, obstacle grids are filled with black, and non-obstacle grids are filled with white. Thus, obstacle information within the field of view of the second radar can be determined based on the distribution of the black obstacle grids.

[0083] In some embodiments, the first condition is that the grid height is greater than or equal to a first threshold, the height difference is greater than or equal to a second threshold, and the obstacle attribute is a non-obstacle grid.

[0084] When the grid height is greater than or equal to the first threshold, the possibility that the object corresponding to the first grid is located in the ground area in the LiDAR system coordinate system is excluded. When the height difference is greater than or equal to the second threshold, it indicates that the object corresponding to the first grid may be a fixed roadside object or obstacle such as a green belt, tree, utility pole, or traffic light. When the obstacle attribute is an obstacle grid, it indicates that there is an obstacle that the perception algorithm can identify, such as a car, pedestrian, or cyclist, in the area corresponding to the first grid. When the obstacle attribute is a non-obstacle grid, the object corresponding to the first grid may be a fixed roadside object such as a green belt, or the area corresponding to the first grid may be free of obstacles.

[0085] As shown above, the first threshold is used to exclude ground obstacles, and the second threshold is used to distinguish between dynamic obstacles such as pedestrians and vehicles, and static obstacles such as green belts, trees, and steps. Each time the lidar system starts, the processor calls the first and second thresholds pre-stored in the memory. The first and second thresholds are pre-set fixed thresholds. To ensure that the first and second thresholds conform to actual road conditions, in some embodiments, the first and second thresholds are determined based on the point cloud collected by the first radar in the first N-1 acquisition cycles. For example, the first threshold is 30cm, and the range of the second threshold is 30cm to 50cm.

[0086] For example, in such Figure 1In the autonomous driving scenario shown, the first radar scans the area in front of vehicle 200. A perception algorithm is typically used to identify obstacles in the point cloud collected by the first radar in the Nth acquisition cycle for obstacle avoidance or path planning. As vehicle 200 moves forward, the area it traverses is the historically passable road area. Based on the historical frame point cloud corresponding to this historically passable area (i.e., the point cloud collected by the first radar in the previous N-1 acquisition cycles), road conditions, such as ground elevation or ground slope, can be determined. Therefore, a first threshold and a second threshold can be determined based on the road conditions, which helps to more accurately segment the area to which the first grid belongs, making the information observed by the first radar at the grid layer more accurate and improving the targeting and accuracy of obstacle detection.

[0087] In some embodiments, a second radar observation grid layer is obtained based on the point cloud acquired by the second radar during the Nth acquisition cycle, including:

[0088] S214: Rasterize the point cloud acquired by the second radar during the Nth acquisition cycle to obtain a second grid, wherein the second grid is marked with grid height and height difference.

[0089] The rasterization process can be referred to in step S211 for further details. Similarly, the point cloud acquired by the second radar in the Nth acquisition cycle is rasterized, including the following steps: Projecting the point cloud acquired by the second radar in the Nth acquisition cycle onto the xy plane to form a two-dimensional point cloud containing height values. Then, dividing the xy plane containing the two-dimensional point cloud into a grid of size r*r, obtaining the grid layer corresponding to the point cloud acquired by the second radar in the Nth acquisition cycle. Here, r is the same as the r mentioned in step S211. When converting the point cloud acquired by the three-dimensional second radar in the Nth acquisition cycle into a two-dimensional point cloud, the height value of each laser point in the z-axis direction of the point cloud acquired by the second radar in the Nth acquisition cycle is retained as the height value of the two-dimensional point cloud. In one embodiment, the xy plane is a horizontal plane parallel to the vehicle's driving direction.

[0090] The second grid is any grid in the grid layer corresponding to the point cloud acquired by the second radar during the Nth acquisition cycle. The second grid is marked with a grid height and a height difference. It is understood that the two-dimensional point cloud in the second grid includes multiple laser points. In some embodiments, the median height of the point cloud points in the second grid can be used as the grid height of the second grid, and the absolute value of the difference between the maximum and minimum heights of the point cloud points in the second grid can be used as the height difference of the second grid.

[0091] S215: Based on the grid height and height difference of the second grid, determine and mark the ground attributes of the second grid to obtain the second radar observation grid layer.

[0092] Ground attributes include ground grids or non-ground grids. Ground grids are grids that represent the ground, while non-ground grids are grids that represent non-ground elements.

[0093] The grid height and height difference reflect the height and dimensions of the object corresponding to the second grid. Given the significant differences in height and dimensions between the ground and other non-ground objects, the height and dimensions of the object reflected by the second grid are used to determine whether the second grid is a ground grid or a non-ground grid. For any second grid in the grid layer corresponding to the point cloud acquired by the second radar within the Nth acquisition cycle, its ground attributes are determined in the above manner. After the ground attributes of each second grid in the grid layer corresponding to the point cloud acquired by the second radar within the Nth acquisition cycle are determined, the second radar observation grid layer is obtained. Here, the second radar observation grid layer refers to the grid map of the second radar that is currently used for observation and is the most up-to-date in time sequence.

[0094] By using the above method, each second grid cell in the grid layer corresponding to the point cloud acquired by the second radar in the Nth acquisition cycle is marked with ground attributes, so that the obtained second radar observation grid layer can distinguish between ground and non-ground, which is beneficial for subsequent accurate perception of obstacles in the field of view of the second radar.

[0095] In some embodiments, determining and marking the ground attributes of the second grid based on the grid height and height difference of the second grid to obtain a second radar observation grid layer includes: determining and marking the second grid that meets the second condition as a non-ground grid, and determining and marking the second grid that does not meet the second condition as a ground grid.

[0096] In some embodiments, a value of 1 can be used to represent a non-ground grid and a value of 0 can be used to represent a ground grid. That is, when the ground attribute of the second grid is 1, it represents that it is a non-ground grid; when the ground attribute of the second grid is 0, it represents that it is a ground grid.

[0097] In some embodiments, the second condition is that the grid height is greater than or equal to a third threshold, or that the height difference is greater than or equal to a fourth threshold.

[0098] When the grid height is greater than or equal to the third threshold, it indicates that the object corresponding to the second grid has a high height in the lidar system coordinate system, and may be an obstacle or other non-ground object. Therefore, the second grid that satisfies the condition of a grid height greater than or equal to the third threshold is marked as a non-ground grid. In some embodiments, the third threshold can be a pre-set fixed threshold, such as 30cm or 40cm. When the height difference is greater than or equal to the fourth threshold, it indicates that the object corresponding to the second grid has a certain height difference and cannot be ground. Therefore, the second grid that satisfies the condition of a height difference greater than or equal to the fourth threshold is marked as a non-ground grid. In some embodiments, the fourth threshold can be a pre-set fixed threshold, such as 20cm or 30cm.

[0099] In some embodiments, the third and fourth thresholds are determined based on the point cloud acquired by the second radar over the first N-1 acquisition cycles. For example, the third threshold is 40 cm and the fourth threshold is 30 cm; in other embodiments, the third threshold is 30 cm and the fourth threshold is 20 cm.

[0100] In some embodiments, the main grid map observation layer is obtained based on the first radar observation grid layer and the second radar observation grid layer, including:

[0101] S221: Overlay the first radar observation grid layer and the second radar observation grid layer to obtain a fused grid map layer.

[0102] In this context, "overlay" refers to stitching together or overlapping the grids in the first radar observation grid layer and the second radar observation grid layer. Specifically, for fields of view where the first and second radars do not overlap, the first grid of the first radar observation grid layer and the second grid of the second radar observation grid layer are stitched together; for fields of view where the first and second radars overlap, the regional attributes of the first grid in the first radar observation grid layer and the ground attributes of the second grid in the second radar observation grid layer can be combined.

[0103] S222: Based on the region to which the first radar observation grid layer belongs and the ground attributes of the second radar observation grid layer, determine and mark the obstacle attributes of the fused grid map layer to obtain the main grid map observation layer.

[0104] Specifically, for the first grid cell in the fused raster map layer, its obstacle attribute is determined based on the region to which the first grid cell belongs. For example, if the first grid cell belongs to the region of interest, it is determined to be an obstacle grid cell; if the first grid cell belongs to the region of non-interest, it is determined to be a non-obstacle grid cell.

[0105] For the second grid cell in the fused raster map layer, its obstacle attributes are determined based on the ground attributes of the second grid cell. For example, if the second grid cell is a ground grid cell, it is determined to be a non-obstacle grid cell; if the second grid cell is a non-ground grid cell, it is determined to be an obstacle grid cell.

[0106] For a third grid cell in a fused grid map (formed by the overlap of the first and second grid cells), the obstacle attribute of the third grid cell is determined based on the region to which the first grid cell belongs and the ground attributes of the second grid cell. In some embodiments, the detection results of the first radar can be trusted. When the first grid cell belongs to the area of ​​interest, the third grid cell is determined to be an obstacle grid cell regardless of whether the second grid cell is a ground grid cell or a non-ground grid cell; when the first grid cell belongs to the area of ​​non-interest, the third grid cell is determined to be a non-obstacle grid cell regardless of whether the second grid cell is a ground grid cell or a non-ground grid cell. In some embodiments, when the first grid cell belongs to the area of ​​interest and the second grid cell is a non-ground grid cell, the third grid cell is determined to be an obstacle grid cell; when the first grid cell belongs to the area of ​​non-interest, the third grid cell is determined to be a non-obstacle grid cell.

[0107] By using the regional attributes of the first radar-observed grid layer and the ground attributes of the second radar-observed grid layer to dually determine the obstacle attributes of the fused grid map layer, the accuracy of obstacle detection is improved.

[0108] Please see Figure 5 , Figure 5 This is a schematic flowchart illustrating an obstacle detection method provided in another embodiment of this application. Figure 5 As shown, method A10 includes the following steps:

[0109] A11: Obtain the point cloud acquired by the first radar and the second radar in N acquisition cycles in sequence, where N is an integer greater than 1.

[0110] A12: Rasterize the point cloud acquired by the first radar during the Nth acquisition cycle to obtain the first grid; based on the obstacle information of the point cloud acquired by the first radar during the Nth acquisition cycle, label the obstacle attributes of the first grid; determine and label the regions to which the first grids that meet the first condition belong as non-interested sensing regions, and determine and label the regions to which the first grids that do not meet the first condition belong as interested sensing regions, thus obtaining the first radar observation grid layer. The first condition is that the grid height is greater than or equal to a first threshold, the height difference is greater than or equal to a second threshold, and the obstacle attribute is a non-obstacle grid.

[0111] The point cloud acquired by the first radar in the Nth acquisition cycle is rasterized to identify fixed roadside objects such as green belts or steps. Based on the obstacle attributes, the observation layer corresponding to the point cloud acquired by the first radar in the Nth acquisition cycle is divided into regions of interest and regions of non-interest. This allows the obtained first radar observation raster layer to exclude fixed roadside objects such as green belts or steps, thereby reducing the interference of these fixed roadside objects on obstacles in the field of view of the second radar.

[0112] A13: Rasterize the point cloud acquired by the second radar during the Nth acquisition cycle to obtain a second grid. Each second grid is marked with its grid height and height difference. Second grids that meet the second condition are identified and marked as non-ground grids, while those that do not meet the second condition are identified and marked as ground grids, thus obtaining the second radar observation layer. The second condition is that the grid height is greater than or equal to a third threshold, or the grid height difference is greater than or equal to a fourth threshold. The third and fourth thresholds are determined based on the point cloud acquired by the second radar during the first N-1 acquisition cycles.

[0113] By identifying ground and non-ground elements in the above manner, the ground attributes are marked on each second grid in the observation layer corresponding to the point cloud acquired by the second radar during the Nth acquisition cycle. This enables the obtained second radar observation grid layer to distinguish between ground and non-ground elements, which helps to improve the targeting and accuracy of obstacle detection.

[0114] A14: Overlay the first radar observation grid layer and the second radar observation grid layer to obtain the fused grid map layer; determine and mark the obstacle attributes of the fused grid map layer according to the region to which the first radar observation grid layer belongs and the ground attributes of the second radar observation grid layer to obtain the main grid map observation layer for the Nth acquisition cycle.

[0115] By using the regional attributes of the first radar-observed grid layer and the ground attributes of the second radar-observed grid layer to dually determine the obstacle attributes of the fused grid map layer, the accuracy and reliability of the detection results can be improved.

[0116] A15: Based on the observation layer of the main grid map in the Nth acquisition cycle and the main grid map in the (N-1)th acquisition cycle, obtain the main grid map for the Nth acquisition cycle.

[0117] A16: Perform cluster analysis on the main grid map of the Nth acquisition cycle to obtain obstacle information within the field of view of the second radar.

[0118] In summary, this application addresses the temporal overlap between the field of view of the first radar and the field of view of the second radar. It employs a main grid map for the (N-1)th acquisition cycle, derived from point clouds collected by the first and second radars in the previous N-1 acquisition cycles, to represent the historical obstacle information sensed by the first radar. Thus, by using the historical obstacle information sensed by the first radar as prior knowledge, and combining it with the point clouds collected by the first and second radars in the Nth acquisition cycle, the main grid map for the Nth acquisition cycle is determined. This ensures that the main grid map for the Nth acquisition cycle reflects the obstacle information within the field of view of the second radar, facilitating accurate identification of obstacles within the second radar's field of view based on this main grid map. Therefore, by using the historical obstacle information sensed by the first radar as prior knowledge, compared to clustering analysis or deep learning detection on the point clouds collected by the second radar in the Nth acquisition cycle, it is possible to accurately perceive obstacles within the second radar's field of view while effectively saving computational resources. On the other hand, by generating the main grid map for the Nth acquisition cycle, the three-dimensional point cloud is reduced to a two-dimensional grid map. Cluster analysis is then performed in the two-dimensional space to identify obstacles, which can effectively reduce the time and space complexity of the recognition algorithm, thereby saving computing power.

[0119] This application also provides a computer-readable storage medium storing a computer program that, when executed by one or more processors, implements the obstacle detection method in any of the above method embodiments.

[0120] In some embodiments, the storage medium may be a ferroelectric random access memory (FRAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EE PROM), flash memory, magnetic surface memory, optical disc, or a compact disc read-only memory (CD-ROM), etc.; or it may be a device that includes one or any combination of the above-mentioned memories.

[0121] In some embodiments, a computer program may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0122] In the description of this application, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Those skilled in the art can understand the specific meaning of these terms in this application based on the specific circumstances. Furthermore, in the description of this application, unless otherwise stated, "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship.

[0123] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this invention. In practical applications, those skilled in the art can select some or all of the workflow to achieve the purpose of this embodiment according to actual needs, and no limitations are imposed here. Furthermore, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0124] The above are merely preferred embodiments of the present invention and do not limit the patent scope of the present invention. Without conflict, the technical features in the different embodiments above can be combined, the steps can be implemented in any order, and many other variations exist as described above regarding different aspects of this application. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features, without departing from the concept of the present application. Any equivalent structural or procedural transformations made based on the content of this specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this invention.

Claims

1. An obstacle detection method applied to a lidar system, the lidar system comprising a first radar and a second radar, characterized in that, The method includes: Acquire point clouds collected by the first and second radars in N acquisition cycles in a sequential order, where N is an integer greater than 1; Based on the point cloud acquired by the first radar in the Nth acquisition cycle, the point cloud acquired by the second radar in the Nth acquisition cycle, and the main grid map in the (N-1)th acquisition cycle, the main grid map for the Nth acquisition cycle is obtained. The main grid map for the (N-1)th acquisition cycle is determined based on the point clouds acquired by the first radar and the second radar in the previous N-1 acquisition cycles. The point cloud acquired by the first radar in the Nth acquisition cycle is marked with obstacle information detected by a perception algorithm. The field of view of the first radar and the field of view of the second radar intersect spatially. The point cloud acquired by the radar reflects obstacle information in the first region of the field of view of the second radar during the Nth acquisition cycle. The first region is the area where the field of view of the second radar and the field of view of the first radar intersect during the Nth acquisition cycle. The field of view of the first radar and the field of view of the second radar overlap temporally. The main grid map of the N-1th acquisition cycle reflects obstacle information in the second region of the field of view of the second radar during the Nth acquisition cycle. The second region is the area where the field of view of the second radar and the field of view of the first radar do not intersect during the Nth acquisition cycle. Cluster analysis is performed on the main grid map of the Nth acquisition cycle to obtain obstacle information within the field of view of the second radar.

2. The method according to claim 1, characterized in that, The process of obtaining the main grid map for the Nth acquisition cycle based on the point cloud acquired by the first radar in the Nth acquisition cycle, the point cloud acquired by the second radar in the Nth acquisition cycle, and the main grid map of the (N-1)th acquisition cycle includes: Based on the point cloud acquired by the first radar during the Nth acquisition cycle, a first radar observation grid layer is obtained; based on the point cloud acquired by the second radar during the Nth acquisition cycle, a second radar observation grid layer is obtained. Based on the first radar observation grid layer and the second radar observation grid layer, the main grid map observation layer for the Nth acquisition cycle is obtained; The main grid map for the Nth acquisition cycle is obtained based on the observation layer of the main grid map for the Nth acquisition cycle and the main grid map for the (N-1)th acquisition cycle.

3. The method according to claim 2, characterized in that, The step of obtaining the first radar observation grid layer based on the point cloud acquired by the first radar during the Nth acquisition cycle includes: The point cloud acquired by the first radar during the Nth acquisition cycle is rasterized to obtain a first grid. The first grid is marked with a grid height and a height difference. The grid height is the median height of the point cloud points in the grid, and the height difference is the absolute value of the difference between the maximum and minimum heights of the point cloud points in the grid. Based on the obstacle information of the point cloud acquired by the first radar during the Nth acquisition cycle, the obstacle attributes of the first grid are marked, wherein the obstacle attributes include obstacle grids or non-obstacle grids. Based on the grid height, height difference, and obstacle attributes of the first grid, the region to which the first grid belongs is determined and marked to obtain the first radar observation grid layer, wherein the region includes a region of interest or a region of non-interest.

4. The method according to claim 3, characterized in that, The process of obtaining the second radar observation grid layer based on the point cloud acquired by the second radar during the Nth acquisition cycle includes: The point cloud acquired by the second radar during the Nth acquisition cycle is rasterized to obtain a second grid, wherein the second grid is marked with the grid height and the height difference; Based on the grid height and height difference of the second grid, the ground attributes of the second grid are determined and marked to obtain the second radar observation grid layer, wherein the ground attributes include ground grids or non-ground grids.

5. The method according to claim 3, characterized in that, The step of determining and marking the area to which the first grid cell belongs based on the grid cell height, height difference, and obstacle attributes includes: The region to which the first grid belongs that satisfies the first condition is determined and marked as the non-interest sensing region, and the region to which the first grid belongs that does not satisfy the first condition is determined and marked as the interest sensing region. The first condition is that the grid height is greater than or equal to a first threshold, the height difference is greater than or equal to a second threshold, and the obstacle attribute is the non-obstacle grid. The first threshold and the second threshold are determined based on the point cloud collected by the first radar in the first N-1 acquisition cycles.

6. The method according to claim 4, characterized in that, The step of determining and marking the ground attributes of the second grid cell based on its grid height and height difference includes: The second grid that meets the second condition is identified and marked as the non-ground grid, and the second grid that does not meet the second condition is identified and marked as the ground grid; The second condition is that the grid height is greater than or equal to the third threshold, or the height difference is greater than or equal to the fourth threshold. The third threshold and the fourth threshold are determined based on the point cloud acquired by the second radar in the first N-1 acquisition cycles.

7. The method according to claim 4, characterized in that, The step of obtaining the main grid map observation layer based on the first radar observation grid layer and the second radar observation grid layer includes: The first radar observation grid layer and the second radar observation grid layer are superimposed to obtain a fused grid map layer; Based on the region to which the first radar observation grid layer belongs and the ground attributes of the second radar observation grid layer, the obstacle attributes of the fused grid map layer are determined and marked to obtain the main grid map observation layer.

8. A lidar system, characterized in that, include: Memory, used to store executable program code; A processor is configured to call and run the executable program code from the memory, causing the lidar system to perform the obstacle detection method as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, implements the obstacle detection method as described in any one of claims 1-7.

Images