Search

CN-121999174-A - Quick point cloud positioning method, equipment and medium based on coding

CN121999174ACN 121999174 ACN121999174 ACN 121999174ACN-121999174-A

Abstract

The invention discloses a quick point cloud positioning method, equipment and a medium based on coding, which comprises the following steps of 1, obtaining a target point cloud, 2, establishing a Z-axis coding mechanism, voxelizing X, Y, Z axes of a point cloud map, binary coding each voxel of the Z axes, generating a plurality of two-dimensional coding matrixes, determining candidate areas, 3, establishing a BFS mode, obtaining candidate poses on the two-dimensional coding matrixes according to the condition that the target point cloud is matched with the source point cloud in the same candidate area in the vertical direction, 4, constructing a search sequence, constructing a Yaw-xy-prz search sequence based on the Yaw angle search range of the Z axes constrained by vehicle motion characteristics, 5, carrying out hierarchical matching search, quickly searching a candidate pose set in the determined candidate areas by using a loose threshold, and recalculating scores to obtain optimal change poses. Therefore, the calculation complexity is reduced, and the positioning searching efficiency and the anti-interference capability are improved.

Inventors

  • GUO XINYANG
  • WU ENGUANG
  • WANG HONGHUI

Assignees

  • 城市之光(深圳)无人驾驶有限公司

Dates

Publication Date
20260508
Application Date
20251231

Claims (10)

  1. 1. The quick point cloud positioning method based on the coding is characterized by comprising the following steps of: Step 1, acquiring a target point cloud through a laser radar; Step 2, a Z-axis coding mechanism is established, a X, Y, Z axis of a point cloud map is subjected to voxelization according to first resolutions of different levels, binary coding is carried out on whether point clouds exist in each voxel in the Z-axis direction, the voxel coding of the point clouds is 1, the voxel coding of the point clouds is 0, a plurality of two-dimensional coding matrixes are generated, the value of each unit in each two-dimensional coding matrix represents the occupation condition in the Z-axis direction, and the two-dimensional coding matrixes of different point cloud frames are compared to screen out candidate areas with similar height structures; Step 3, establishing a BFS mode, calculating the score in the vertical direction of a candidate area according to the number of final marks of 1 in the vertical direction of the candidate area with the same target point cloud and source point cloud on the two-dimensional coding matrix with the first resolution of each level, and obtaining a candidate pose by matching according to the score condition; Step 4, constructing a search sequence, restraining a Yaw angle search range of a Z axis based on the motion characteristics of the ground vehicle, preferentially determining the Yaw angle of the Z axis, and then determining XY position information and PRZ pose parameters to construct a Yaw-XY-PRZ search sequence; And 5, hierarchical matching search, namely quickly searching a candidate pose set matched with the target point cloud in the determined candidate region by using a loose threshold value, and calculating scores by adopting a second resolution based on the obtained candidate pose set to obtain the optimal change pose.
  2. 2. The method for locating a rapid point cloud based on coding of claim 1, wherein in the step 2, when the X, Y, Z axes of the point cloud map are voxelized, each voxel of the Z axis of the same X, Y value of the point cloud map is correspondingly binary coded for completing interval discretization of the point cloud map.
  3. 3. The method for locating the point cloud based on the coding is characterized in that in the step 2, after interval discretization is carried out on the point cloud map, candidate areas with similar height structures are screened out through bit operation and comparison of two-dimensional coding matrixes of different point cloud frames, and the candidate areas are used for completing vertical feature comparison of the two-dimensional coding matrixes.
  4. 4. The method of claim 1, wherein in step 2 ‌, a 64-bit integer is used to store binary codes, and the 64-bit integer is split into a high-order 32 bits and a low-order 32 bits, which are stored in two 32-bit variables respectively, so as to reduce memory occupation and implement fast bit operation.
  5. 5. The method of claim 1, wherein step 3 includes performing an and operation on the number of final marks 1 in the vertical direction of the candidate areas of the target point cloud and the source point cloud, and adding the number of final marks 1 to the result of the and operation to calculate a score, and if the score is lower than a corresponding preset threshold, performing dynamic threshold pruning and terminating branching for reducing ineffective searching ‌.
  6. 6. The code-based rapid point cloud positioning method of claim 5, wherein priority queue management is performed according to score conditions, the expansion nodes are ordered according to the scores, and high-score candidate poses in the candidate areas are preferentially processed for avoiding repeated computation ‌.
  7. 7. The method for locating a fast point cloud based on coding according to claim 6, wherein in step 3, the corresponding scores of the two-bit coding matrices of the first resolution of the same hierarchy are stored in a priority queue, and the first to fifth scores are preset from high score to low score, and pruning is performed below the third score for optimizing the calculation range of the same hierarchy.
  8. 8. The encoding-based rapid point cloud positioning method as claimed in any of claims 1 to 7, wherein in step 5, said loose threshold used by said candidate region includes at least a rotation, translation, feature matching threshold, which has a value ranging from 0.05 to 0.5m; Based on the candidate pose set acquired under the loose threshold condition, the average point distance set by the second resolution is smaller than the average point distance set by the first resolution, and the value range of the average point distance of the second resolution is 0.5-2cm.
  9. 9. An electronic device comprising a memory storing executable program code, a processor coupled to the memory, the processor invoking the executable program code stored in the memory for performing the encoding-based rapid point cloud positioning method of any of claims 1-8.
  10. 10. A computer readable storage medium, characterized in that the computer readable storage medium stores a computer program, wherein the computer program causes a computer to execute the encoding-based rapid point cloud positioning method according to any one of claims 1 to 8.

Description

Quick point cloud positioning method, equipment and medium based on coding Technical Field The invention relates to the technical field of automatic driving positioning, in particular to a rapid point cloud positioning method, device and medium based on coding. Background The automatic driving technology is a core technology development trend in the intelligent traffic field, and particularly the positioning accuracy of automatic driving directly influences the safety and reliability of a vehicle in the driving process. The point cloud positioning in automatic driving refers to a technology of determining the accurate position and orientation of a vehicle in the environment by matching three-dimensional point cloud data acquired by a sensor such as a laser radar (LiDAR) with a pre-built high-precision point cloud map. In the prior art, automatic driving generally depends on effective Global Navigation Satellite System (GNSS) data, projects a point cloud frame acquired by a current laser radar into a pre-constructed point cloud map coordinate system, and then realizes high-precision positioning based on a point cloud map, and searches and positions by adopting six degrees of freedom sequences of XYZ-RPY of a fixed coordinate system (world coordinate system), wherein the rotation sequence of the fixed coordinate system is about an X-axis Roll, a Y-axis Pitch and a Z-axis Yaw, and belongs to an RPY angle representation method. However, this prior art still has the following problems: 1. The anti-interference capability and the positioning speed are insufficient, in practical application, GNSS signals are easy to be interfered by factors such as shielding or magnetic fields, multipaths and the like under special environments (urban canyons, tunnels, underground parking lots and the like), so that the positioning signals are attenuated, delayed or even interrupted, and therefore an automatic driving system cannot acquire positioning information through the GNSS signals rapidly and with high precision, the autonomous navigation capability of a vehicle is further influenced, and even traffic safety accidents can be caused. 2. The method has the advantages that the calculation complexity is high, the searching efficiency is low, the point cloud matching in the prior art has obvious limitation, the point cloud matching is used as positioning supplement in the GNSS failure scene, the real-time point cloud and the map point cloud are used for accurately matching to determine the vehicle position, the large-scale feature extraction, matching and optimization calculation processes are required to be carried out on the point cloud data, the calculation complexity is extremely high, the matching range is wide, the searching space is huge, the positioning searching efficiency is seriously affected, and the current high-quality development requirement on automatic driving cannot be met. Disclosure of Invention In order to overcome the defects of the prior art, one of the purposes of the invention is to provide a rapid point cloud positioning method, device and medium based on coding. One of the purposes of the invention is realized by adopting the following technical scheme that the quick point cloud positioning method based on the coding comprises the following steps: Step 1, acquiring a target point cloud through a laser radar; Step 2, a Z-axis coding mechanism is established, a X, Y, Z axis of a point cloud map is subjected to voxelization according to first resolutions of different levels, binary coding is carried out on whether point clouds exist in each voxel in the Z-axis direction, the voxel coding of the point clouds is 1, the voxel coding of the point clouds is 0, a plurality of two-dimensional coding matrixes are generated, the value of each unit in each two-dimensional coding matrix represents the occupation condition in the Z-axis direction, and the two-dimensional coding matrixes of different point cloud frames are compared to screen out candidate areas with similar height structures; Step 3, establishing a BFS mode, calculating the score in the vertical direction of a candidate area according to the number of final marks of 1 in the vertical direction of the candidate area with the same target point cloud and source point cloud on the two-dimensional coding matrix with the first resolution of each level, and obtaining a candidate pose by matching according to the score condition; Step 4, constructing a search sequence, restraining a Yaw angle search range of a Z axis based on the motion characteristics of the ground vehicle, preferentially determining the Yaw angle of the Z axis, and then determining XY position information and PRZ pose parameters to construct a Yaw-XY-PRZ search sequence; And 5, hierarchical matching search, namely quickly searching a candidate pose set matched with the target point cloud in the determined candidate region by using a loose threshold value, and calculating scores by adopting