Obstacle detection method and device based on laser radar

Abstract

The present invention provides a laser radar-based obstacle detection method and device, comprising: utilizing spatial hashing to columnarize acquired point cloud data to obtain a plurality of voxel grids; obtaining target features corresponding to each of the plurality of voxel grids based on the plurality of voxel grids; and performing bi-clustering on the target features to obtain obstacle detection results. The present invention utilizes bi-clustering of columnar voxels to obtain obstacle detection, thereby improving clustering efficiency and the accuracy of obstacle detection. Columnarizing the acquired point cloud data using spatial hashing rapidly obtains voxel grids and ensures that each voxel grid is of the same size, thereby facilitating subsequent bi-clustering. Furthermore, this method reduces the space required for point cloud data and facilitates rapid point cloud data retrieval.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to a laser radar-based obstacle detection method and device. Background Art

[0002] Autonomous vehicles (also known as self-piloting automobiles), also known as driverless cars, are a technology that uses computer-controlled devices to achieve autonomous driving on the road. The realization of autonomous driving relies on the coordinated efforts of artificial intelligence, visual computing, radar, and positioning components. Due to the complex road conditions and the numerous obstacles such as pedestrians and vehicles, the key to autonomous driving is to identify obstacles and then plan routes to avoid them.

[0003] At present, most general obstacle detection algorithms based on lidar can realize obstacle recognition. General obstacle detection algorithms mainly fit the ground equation based on random consensus sampling (RANSAC) on the original point cloud data, then apply a threshold in the vertical direction of the ground to distinguish ground points from non-ground points, and then identify non-bottom surface points to obtain obstacle information.

[0004] Since general obstacle detection algorithms directly process point clouds in units of points, the algorithm's efficiency depends on the number of points. LiDAR is now shifting from high-cost mechanical types to low-cost, mass-producible solid-state types. Compared with mechanical LiDAR, solid-state LiDAR generates denser point clouds and more points, thereby increasing the time efficiency of general obstacle detection. Summary of the Invention

[0005] The present invention provides a method and device for obstacle detection based on laser radar, which is used to solve the defect of low obstacle detection efficiency due to dense point cloud in the prior art.

[0006] The present invention provides an obstacle detection method based on laser radar, comprising: performing columnar voxelization on acquired point cloud data using spatial hashing to obtain a plurality of voxel grids; obtaining target features corresponding to each of the plurality of voxel grids based on the plurality of voxel grids; and performing double clustering on the target features to obtain obstacle detection results.

[0007] According to a lidar-based obstacle detection method provided by the present invention, the dual clustering of the target features includes: performing a first clustering based on the target features in units of the voxel grid to obtain a clustering result; obtaining an obstacle pixel cluster based on the clustering result; and performing a second clustering of each obstacle pixel within the obstacle pixel cluster based on position coordinates to obtain an obstacle detection result.

[0008] According to a lidar-based obstacle detection method provided by the present invention, after the target features are bi-clustered, the method further includes: obtaining point cloud coordinates corresponding to each voxel square in each cluster based on the obstacle detection results; and obtaining obstacle instance information based on the point cloud coordinates corresponding to each voxel square in each cluster.

[0009] According to a lidar-based obstacle detection method provided by the present invention, before the target features are double-clustered, the method further includes: based on a preset threshold, screening the target features to remove target features corresponding to non-obstacles from the target features.

[0010] According to a laser radar-based obstacle detection method provided by the present invention, obtaining target features corresponding to each voxel square based on the multiple voxel squares includes: screening the multiple voxel squares based on whether the multiple voxel squares contain point cloud data to obtain voxel squares including point cloud data; obtaining target features corresponding to each voxel square including point cloud data based on the voxel squares including point cloud data, the target features including at least one of the elevation difference, average elevation, elevation variance, three-dimensional surface curvature and average reflectivity corresponding to the voxel square including point cloud data.

[0011] According to a laser radar-based obstacle detection method provided by the present invention, before the point cloud data obtained is columnarized using spatial hashing, the method further includes: obtaining laser radar point cloud data; and performing range filtering on the point cloud data based on a region of interest.

[0012] The present invention also provides an obstacle detection device based on laser radar, comprising: a data processing module, which uses spatial domain hashing to columnarize the acquired point cloud data to obtain multiple voxel grids; a feature acquisition module, which obtains target features corresponding to each of the multiple voxel grids based on the multiple voxel grids; and an obstacle detection module, which performs double clustering on the target features to obtain obstacle detection results.

[0013] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any of the above-described laser radar-based obstacle detection methods are implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described laser radar-based obstacle detection methods.

[0015] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of any of the above-described laser radar-based obstacle detection methods.

[0016] The laser radar-based obstacle detection method and device provided by the present invention perform dual clustering of columnar voxels to obtain obstacle detection, thereby improving clustering efficiency and the accuracy of obstacle detection; the acquired point cloud data is columnarized by spatial domain hashing to quickly obtain voxel grids and ensure that each voxel grid is of the same size, thereby facilitating subsequent dual clustering. In addition, it can also reduce the space of point cloud data and facilitate rapid search of point cloud data. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 1 is a flow chart of the obstacle detection method based on laser radar provided by the present invention;

[0019] Figure 2 Schematic diagram of the structure of the obstacle detection device based on laser radar provided by the present invention;

[0020] Figure 3 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0021] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0022] Figure 1 A schematic flow chart of a laser radar-based obstacle detection method of the present invention is shown, and the method includes:

[0023] S01, using spatial hashing to perform columnar voxelization on the acquired point cloud data to obtain multiple voxel grids;

[0024] S02, obtaining target features corresponding to each voxel grid based on the plurality of voxel grids;

[0025] S03, performing double clustering on the target features to obtain the obstacle detection result.

[0026] It should be noted that S0N in this specification does not represent the order of the obstacle detection method based on laser radar. The obstacle detection method based on laser radar of the present invention is described in detail below.

[0027] In step S01 , the acquired point cloud data is voxelized using spatial hashing to obtain a plurality of voxel grids.

[0028] In this embodiment, the acquired point cloud data is columnarized by spatial hashing, including: defining a detection space; based on the detection space, columnarizing the acquired point cloud data by using spatial hashing to obtain a plurality of voxel grids. Specifically:

[0029] First, define the detection space, that is, a cube with a set length and width as the bottom and a set height as the detection space. It should be noted that in the actual field related to autonomous driving, it is necessary to pay attention to general obstacles in the height direction. General obstacles can be any obstacles that may appear on the road surface of a closed road and require driving to brake or avoid them. They include not only motorized / non-motorized vehicles and pedestrians, but also roadblocks, tripod warning signs, tires, scattered car parts, etc. Therefore, the set height can be set according to actual needs and is not further limited here. In addition, the length and width of the detection space can be set according to the parameters of the lidar and the actual road conditions, and are not further limited here.

[0030] For example, if the set height reference value of the detection space is 2 meters, based on the data collection characteristics of the current vehicle and the lidar installed on the current vehicle, one of the length and width of the detection space should pass through the location of the current vehicle.

[0031] Next, based on the detection space, spatial hashing is used to columnarize the acquired point cloud data, resulting in multiple voxel grids. In other words, spatial hashing is used to segment the detection space, resulting in multiple voxel grids. It should be noted that to prevent overlap between the projections of the individual voxel grids on the ground plane, the height of the voxel grids can be set to the same height as the detection space. Furthermore, spatial hashing is used to segment the length and width of the detection space, converting the point cloud data into multiple voxel grids of fixed length and width.

[0032] In this embodiment, voxel is the abbreviation of volume pixel, which is the smallest unit divided in the digital three-dimensional space.

[0033] In an optional embodiment, before performing columnar voxelization on the acquired point cloud data using spatial hashing, the method further includes: acquiring lidar point cloud data. It should be noted that the lidar point cloud data is acquired by detecting a lidar installed on a vehicle. Point cloud data refers to data that records scanning information in the form of points. Each point cloud data acquired by the lidar scan contains a three-dimensional coordinate (X, Y, Z) and a laser reflection intensity information (Intensity).

[0034] In an optional embodiment, in order to reduce the amount of point cloud data and improve the efficiency of subsequent obstacle detection, after acquiring the lidar point cloud data, the point cloud data can be range filtered based on the region of interest (ROI). It should be noted that the region of interest is set as the rectangular range of obstacles that may affect driving in the vehicle's forward lane [x min ,y min , z nin ,x max ,y max ,z max ] to filter out the point cloud data of the region of interest (ROI), so as to facilitate the subsequent columnar voxelization of the point cloud data of the region of interest (ROI), reduce the amount of data, and improve the speed of obstacle detection. The forward lane can be a single lane or a two-way lane.

[0035] In an optional embodiment, to further improve the accuracy of obstacle detection, after performing range filtering on the point cloud data based on the region of interest (ROI), the LiDAR point cloud data can also be pre-processed to effectively filter out clutter interference from individual radar points. Specifically, median filtering is used to smooth the LiDAR point cloud data to remove isolated noise points. Isolated noise points include points that display a distance of 0 due to undetectable echoes.

[0036] Step S02 : obtaining target features corresponding to each voxel grid based on the plurality of voxel grids.

[0037] It should be noted that, by obtaining the target features corresponding to each voxel square, it is convenient to subsequently perform bi-clustering on each voxel square according to the target features.

[0038] In an optional embodiment, to improve the efficiency of bi-clustering, when obtaining target features corresponding to each voxel square based on multiple voxel squares, the following steps are performed: filtering the multiple voxel squares based on whether they contain point cloud data to obtain voxel squares containing point cloud data; and obtaining target features corresponding to each voxel square containing point cloud data based on the voxel squares containing point cloud data, wherein the target features include at least one of the elevation difference, average elevation, elevation variance, three-dimensional surface curvature, and average reflectivity corresponding to the voxel square containing point cloud data. It should be noted that by filtering the voxel squares to remove voxel squares that do not contain point cloud data, the number of voxel squares required for bi-clustering is reduced, thereby further improving the efficiency of bi-clustering. In addition, voxel squares containing point cloud data can be understood as non-empty voxel squares. The corresponding features can be selected to constitute the target features based on actual computational requirements and are not further limited here.

[0039] Specifically, the elevation difference is expressed as:

[0040]

[0041] in, represents the elevation difference of the points contained in the i-th non-empty voxel grid, Indicates the maximum elevation, Indicates the minimum elevation.

[0042] Mean elevation, expressed as:

[0043]

[0044] in, represents the average elevation, k represents the number of points in the non-empty voxel grid, Indicates the elevation value of the jth point contained in the i-th non-empty voxel grid.

[0045] Elevation variance, expressed as:

[0046]

[0047] in, represents the elevation variance, k represents the number of points in the i-th non-empty voxel grid, represents the elevation value of the jth point in the i-th non-empty voxel grid, Indicates the average elevation.

[0048] The three-dimensional surface curvature σ satisfies:

[0049]

[0050] Where λ0, λ1, and λ2 are the three eigenvectors of the covariance matrix C, and λ0<λ1<λ2; the covariance matrix C satisfies:

[0051]

[0052] Where k is the number of detection point cloud data in the voxel grid; p i is the coordinate of the i-th detection point cloud data in the voxel grid; is the mean value of the coordinates of the detection point cloud data within the voxel grid.

[0053] The average reflectivity is expressed as:

[0054]

[0055] in, represents the average reflectivity, k represents the number of points in the i-th non-empty voxel grid, Represents the reflectivity of the j-th point contained in the i-th non-empty voxel square.

[0056] Step S03: perform double clustering on the target features to obtain obstacle detection results.

[0057] It should be noted that the double clustering of target features includes: performing a first clustering based on the target features in voxel grids to obtain a clustering result; obtaining an obstacle pixel cluster based on the clustering result; and performing a second clustering of each obstacle pixel in the obstacle pixel cluster based on the position coordinates to obtain an obstacle detection result.

[0058] In this embodiment, first, a first clustering is performed based on the target features in voxel grids to obtain a clustering result. Specifically, the first clustering is performed based on the target features in voxel grids, including: performing similarity calculation on the target features based on a distance function to obtain a clustering result. It should be noted that the clustering result includes ground voxel clusters and obstacle voxel clusters, and each cluster includes multiple similar target features. When performing similarity calculation, a similarity calculation method such as Euclidean distance or cosine similarity can be used to achieve clustering, thereby detecting obstacle voxel grids in non-empty voxel grids and clustering them into a cluster to facilitate subsequent further detection of the obstacle voxel clusters, reduce the amount of data to be detected, and improve detection speed and accuracy.

[0059] Secondly, obstacle voxel clusters are obtained based on the clustering results. It should be noted that when the first clustering is performed based on the target features in the voxel grid, the clustering results obtained include ground voxel clusters and obstacle voxel clusters. Therefore, when obtaining obstacle voxel clusters based on the clustering results, a threshold is pre-set, and voxel clusters with target features greater than the threshold are extracted from the clustering results to obtain obstacle voxel clusters. This facilitates subsequent obstacle identification of the obstacle voxel clusters, saves computing power for obstacle detection, and improves the accuracy of obstacle detection.

[0060] Finally, a second clustering is performed on each obstacle voxel within the obstacle cluster based on the position coordinates to obtain the obstacle detection result. It should be noted that after the second clustering is performed on each obstacle voxel within the obstacle cluster based on the position coordinates, the obtained obstacle detection result includes multiple clusters of voxel squares, and each cluster of voxel squares represents a type of obstacle, which facilitates the subsequent acquisition of corresponding obstacle instance information based on each cluster of voxel squares.

[0061] In an optional embodiment, to improve the efficiency of the first clustering, before performing the secondary clustering on the target features, the method further includes: filtering the target features based on a preset threshold to remove target features corresponding to non-obstacles. It should be noted that the threshold is set based on the characteristics of common obstacles to eliminate non-obstacle point cloud data other than common obstacles, thereby reducing the computational complexity of clustering and improving clustering efficiency and accuracy.

[0062] In an optional embodiment, after performing bi-clustering on the target features, the method further includes: obtaining point cloud coordinates corresponding to each voxel square within each cluster based on the obstacle detection results; and obtaining obstacle instance information based on the point cloud coordinates corresponding to each voxel square within each cluster. It should be noted that the obstacle instance information includes the 3D position and size information of the obstacle instance.

[0063] Among them, the 3D position of the obstacle instance is expressed as (x center ,y center ,z center ),in:

[0064]

[0065]

[0066]

[0067] represents the 3D position of the obstacle instance corresponding to the i-th cluster voxel grid, represents the maximum x-axis coordinate in the point cloud data of the i-th cluster voxel grid, represents the minimum x-axis coordinate of the point cloud data in the i-th cluster voxel grid, represents the maximum y-axis coordinate in the point cloud data of the i-th cluster voxel grid, represents the minimum y-axis coordinate of the point cloud data in the i-th cluster voxel grid, represents the maximum z-axis coordinate in the point cloud data of the i-th cluster voxel grid, represents the minimum z-axis coordinate in the point cloud data of the i-th cluster of voxel squares. It should be noted that a 3D vehicle coordinate system is required. In this embodiment, the forward direction of the vehicle can be used as the positive x-axis direction, the left direction of the vehicle when the ground plane is perpendicular to the x-axis, and the positive z-axis direction is the direction perpendicular to the ground plane and upward.

[0068] In summary, the embodiments of the present invention perform double clustering through columnar voxels to obtain obstacle detection, so as to improve clustering efficiency and improve the accuracy of obstacle detection; the acquired point cloud data is columnarized through spatial hashing to quickly obtain voxel grids and ensure that each voxel grid has the same size, thereby facilitating subsequent double clustering. In addition, the space of point cloud data can also be reduced, which is convenient for rapid search of point cloud data.

[0069] The following describes the laser radar-based obstacle detection device provided by the present invention. The laser radar-based obstacle detection device described below and the laser radar-based obstacle detection method described above can refer to each other.

[0070] Figure 2 The figure shows a schematic structural diagram of an obstacle detection device based on laser radar, which includes:

[0071] The data processing module 21 performs columnar voxelization on the acquired point cloud data using spatial hashing to obtain a plurality of voxel grids;

[0072] A feature acquisition module 22 obtains target features corresponding to each voxel grid based on the plurality of voxel grids;

[0073] The obstacle detection module 23 performs double clustering on the target features to obtain obstacle detection results.

[0074] In this embodiment, the data processing module 21 includes: a setting unit for defining a detection space; a segmentation unit for performing columnar voxelization on the acquired point cloud data using spatial hashing based on the detection space to obtain a plurality of voxel grids.

[0075] It should be noted that when setting the unit to define the detection space, a cube with a set length and width as the bottom and a set height is used as the detection space. Since in the actual field related to autonomous driving, it is necessary to pay attention to general obstacles in the height direction, general obstacles can be any obstacles that may appear on the road surface of a closed road and require driving to brake or avoid them, including not only motorized / non-motorized vehicles and pedestrians, but also roadblocks, tripod warning signs, tires, scattered car parts, etc. Therefore, the set height can be set according to actual needs and is not further limited here. In addition, the length and width of the detection space can be set according to the parameters of the lidar and the actual road conditions, and are not further limited here.

[0076] Furthermore, when the segmentation unit divides the detection space into multiple voxel grids, the height of the voxel grid can be set to the same height as the detection space to prevent overlap of the projections of the individual voxel grids on the ground plane. Furthermore, spatial hashing is used to segment the length and width of the detection space to convert the point cloud data into multiple voxel grids with fixed length and width.

[0077] In an optional embodiment, to facilitate the acquisition of point cloud data, the device further includes a data processing module for acquiring LiDAR point cloud data. It should be noted that LiDAR point cloud data is acquired by a LiDAR installed on a vehicle. Point cloud data refers to data that records scan information in the form of points. Each point cloud data acquired by a LiDAR scan contains a three-dimensional coordinate (X, Y, Z) and a laser reflection intensity information (Intensity).

[0078] In an optional embodiment, to reduce the amount of point cloud data and improve the efficiency of subsequent obstacle detection, the device further includes a filtering module that performs range filtering on the point cloud data based on regions of interest (ROIs). In actual processing, after the data processing module acquires the lidar point cloud data, the filtering module performs range filtering on the point cloud data based on the regions of interest (ROIs).

[0079] It should be noted that the region of interest is set as the rectangular range where the vehicle is pulled forward and may encounter obstacles that may affect driving [x min ,y min , z min , x max ,y max , z max ] to filter out the point cloud data of the region of interest (ROI), thereby facilitating the subsequent columnar voxelization of the point cloud data in the region of interest (ROI), reducing the amount of data and improving the speed of obstacle detection.

[0080] In an optional embodiment, to further improve the accuracy of obstacle detection results, the device also includes a data preprocessing module that preprocesses the LiDAR point cloud data to effectively filter out clutter interference from individual radar points. Specifically, the data preprocessing module includes a smoothing unit that uses median filtering to smooth the LiDAR point cloud data to remove isolated noise points, where isolated noise points include points that display a distance of 0 due to the inability to detect an echo. In the actual processing process, after the filtering module performs range filtering on the point cloud data based on the region of interest (ROI), the data preprocessing module is used to preprocess the LiDAR point cloud data to effectively filter out clutter interference from individual radar points.

[0081] The feature acquisition module 22 includes: a screening unit for screening multiple voxel squares based on whether they contain point cloud data to obtain voxel squares containing point cloud data; and a feature acquisition unit for obtaining target features corresponding to each voxel square containing point cloud data based on the voxel squares containing point cloud data. The target features include at least one of the elevation difference, average elevation, elevation variance, three-dimensional surface curvature, and average reflectivity corresponding to the voxel squares containing point cloud data. It should be noted that the screening unit screens the voxel squares to remove voxel squares that do not contain point cloud data, thereby reducing the number of voxel squares required for biclustering and further improving the efficiency of biclustering. In addition, voxel squares containing point cloud data can be understood as non-empty voxel squares. The corresponding features can be selected to constitute the target features based on actual computational requirements and are not further defined herein.

[0082] The obstacle detection module 23 includes a first clustering unit that performs a first clustering operation based on target features, using voxel grids as units, to obtain clustering results; an obstacle pixel acquisition unit that obtains obstacle pixel clusters based on the clustering results; and a second clustering unit that performs a second clustering operation on each obstacle pixel within the obstacle pixel clusters based on their position coordinates to obtain obstacle detection results. It should be noted that the obtained obstacle detection results include multiple clusters of voxel grids, each representing a specific obstacle, facilitating subsequent acquisition of obstacle instance information based on each cluster.

[0083] Specifically, the first clustering unit includes: a similarity calculation subunit, which performs similarity calculation on the target features based on the distance function to obtain a clustering result. It should be noted that the clustering result includes ground voxel clusters and obstacle voxel clusters, and each cluster includes multiple similar target features. When performing similarity calculation, similarity calculation methods such as Euclidean distance or cosine similarity can be used to achieve clustering, thereby detecting obstacle voxel squares in non-empty voxel squares and clustering them into a cluster to facilitate subsequent further detection of obstacle voxel clusters, reduce the amount of data to be detected, and improve detection speed and accuracy.

[0084] The obstacle pixel acquisition unit includes: a threshold setting subunit that pre-sets a threshold; and a data extraction subunit that extracts voxel clusters whose target features are greater than the threshold from the clustering results to obtain obstacle pixel clusters. This facilitates subsequent obstacle recognition of the obstacle pixel clusters, saves computing power for obstacle detection, and improves the accuracy of obstacle detection.

[0085] In an optional embodiment, to improve the efficiency of the first clustering, the device further includes a screening module that screens target features based on a preset threshold value, removing target features corresponding to non-obstacles. In actual use, before the obstacle detection module 23 performs dual clustering on the target features, the target features are screened to remove non-obstacle point cloud data other than general obstacles, thereby reducing the computational complexity of clustering and improving clustering efficiency and accuracy.

[0086] In an optional embodiment, the device further includes an obstacle instance information determination module, specifically comprising: a point cloud coordinate acquisition unit for obtaining the point cloud coordinates corresponding to each voxel square within each cluster based on the obstacle detection results; and an information calculation unit for obtaining obstacle instance information based on the point cloud coordinates corresponding to each voxel square within each cluster. It should be noted that the obstacle instance information includes the 3D position and size information of the obstacle instance.

[0087] Figure 3 An example of a physical structure diagram of an electronic device is shown below. Figure 3 As shown, the electronic device may include: a processor 31, a communications interface 32, a memory 33, and a communications bus 34. The processor 31, communications interface 32, and memory 33 communicate with each other via the communications bus 34. The processor 31 may call logic instructions in the memory 33 to execute a lidar-based obstacle detection method, which includes: using spatial hashing to columnarize the acquired point cloud data to obtain multiple voxel grids; obtaining target features corresponding to each voxel grid based on the multiple voxel grids; and performing bi-clustering on the target features to obtain obstacle detection results.

[0088] In addition, the logic instructions in the above-mentioned memory 33 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0089] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the lidar-based obstacle detection method provided by the above-mentioned methods, which includes: using spatial hashing to columnar voxelize the acquired point cloud data to obtain multiple voxel grids; based on the multiple voxel grids, obtaining target features corresponding to each voxel grid; and performing double clustering on the target features to obtain obstacle detection results.

[0090] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it is implemented to execute the lidar-based obstacle detection method provided by the above-mentioned methods, the method comprising: using spatial hashing to columnar voxelize the acquired point cloud data to obtain a plurality of voxel grids; based on the plurality of voxel grids, obtaining target features corresponding to each voxel grid; and performing double clustering on the target features to obtain obstacle detection results.

[0091] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0092] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A laser radar-based obstacle detection method, characterized in that: include: The acquired point cloud data is voxelized using spatial hashing to obtain multiple voxel grids; Obtaining target features corresponding to each of the voxel squares according to the plurality of voxel squares; Performing bi-clustering on the target features to obtain obstacle detection results; The performing bi-clustering on the target features includes: Taking the voxel grid as a unit, performing a first clustering based on the target feature to obtain a clustering result; Obtaining obstacle pixel clusters according to the clustering results; Performing a second clustering on each obstacle pixel in the obstacle pixel cluster based on the position coordinates to obtain an obstacle detection result; Taking the voxel grid as a unit, performing a first clustering based on the target feature to obtain a clustering result, including: The target features are similarly calculated based on a distance function to obtain a clustering result, where the clustering result includes a ground voxel cluster and an obstacle voxel cluster.

2. The obstacle detection method based on laser radar according to claim 1, characterized in that: After performing bi-clustering on the target features, the method further includes: According to the obstacle detection results, the point cloud coordinates corresponding to each voxel square in each cluster are obtained; Obstacle instance information is obtained based on the point cloud coordinates corresponding to each voxel grid in each cluster.

3. The obstacle detection method based on laser radar according to claim 1, characterized in that: Before performing bi-clustering on the target features, the method further includes: Based on a preset threshold, the target features are screened to remove target features corresponding to non-obstacles from the target features.

4. The obstacle detection method based on laser radar according to claim 1, characterized in that: Obtaining target features corresponding to each of the voxel grids according to the plurality of voxel grids includes: screening the plurality of voxel squares according to whether the plurality of voxel squares contain point cloud data to obtain voxel squares containing point cloud data; According to the voxel grids including point cloud data, target features corresponding to each voxel grid including point cloud data are obtained, and the target features include at least one of the elevation difference, average elevation, elevation variance, three-dimensional surface curvature and average reflectivity corresponding to the voxel grid including point cloud data.

5. The obstacle detection method based on laser radar according to claim 1, characterized in that: Before performing columnar voxelization on the acquired point cloud data using spatial hashing, the method further includes: Get lidar point cloud data; Range filtering is performed on the point cloud data based on a region of interest.

6. An obstacle detection device based on laser radar, characterized in that: include: The data processing module uses spatial hashing to perform columnar voxelization on the acquired point cloud data to obtain multiple voxel grids; A feature acquisition module, which obtains a target feature corresponding to each voxel grid according to the plurality of voxel grids; The obstacle detection module performs bi-clustering on the target features to obtain obstacle detection results; The obstacle detection module includes: A first clustering unit, using the voxel grid as a unit, performs first clustering based on the target feature to obtain a clustering result; An obstacle pixel acquisition unit obtains obstacle pixel clusters according to the clustering result; a second clustering unit, performing second clustering on each obstacle pixel in the obstacle pixel cluster based on the position coordinates to obtain an obstacle detection result; The first clustering unit includes: The similarity calculation subunit performs similarity calculation on the target features based on a distance function to obtain a clustering result, wherein the clustering result includes a ground voxel cluster and an obstacle voxel cluster.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the laser radar-based obstacle detection method according to any one of claims 1 to 5 are implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the laser radar-based obstacle detection method according to any one of claims 1 to 5 are implemented.

9. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the laser radar-based obstacle detection method according to any one of claims 1 to 5 are implemented.

Images