Search

CN-121994210-A - Dynamic map construction and block differentiation updating method

CN121994210ACN 121994210 ACN121994210 ACN 121994210ACN-121994210-A

Abstract

The invention discloses a dynamic map construction and block differentiation updating method, which divides a map into four types of map blocks, namely stable blocks, slow change blocks, low dynamic blocks or ground blocks, and endows each type of map block with different positioning weights and updating strategies, and dynamically adjusts the type of the map block through a change rate detection algorithm so as to realize self-adaptive map management. The invention improves the robustness of positioning registration, reduces the interference of dynamic objects, realizes the classified management and differential updating of map blocks, ensures the long-term reliability and accuracy of the map, and simultaneously supports the dynamic loading and efficient updating of an infinite map.

Inventors

  • XIAN KAIYI
  • LIU DUO

Assignees

  • 重庆大学

Dates

Publication Date
20260508
Application Date
20260408

Claims (10)

  1. 1. The dynamic map construction and block differentiation updating method is characterized by comprising the following steps of: s1, dividing a global map into a plurality of map blocks according to a fixed space size, and removing a high-dynamic target from the SLAM front end in real time by a moving object detection method; s2, automatically classifying each map block into a stable block, a slow change block, a low dynamic block or a ground block; s3, initializing and generating a map according to the original laser point cloud; S4, positioning and map registration are carried out on the map generated by initialization, and different weights are given to map blocks of different types; S5, calculating the change rate of the map blocks which are registered, and dynamically adjusting the type of the map blocks according to the mean value and the variance of the change rate for a plurality of times in a time window; And S6, updating the stable blocks, the slow change blocks, the low dynamic blocks and the ground blocks by adopting different updating modes respectively, so as to realize differential updating of map blocks of different types.
  2. 2. The method for updating the dynamic map construction and partitioning differentiation according to claim 1, wherein each map block obtained by dividing in the step S1 independently stores point cloud data, statistical characteristics and dynamic state parameters.
  3. 3. The method for dynamic map construction and block differentiation updating according to claim 1, wherein the map blocks divided in step S1 support on-demand loading and unloading to achieve local access.
  4. 4. The method for updating the dynamic map construction and block differentiation according to claim 1, wherein the stable blocks in the step S2 are unchanged, the slow-change blocks are continuously and slowly changed, the low-dynamic blocks exist within a certain time scale and are obviously changed, the ground blocks are obtained by a ground extraction algorithm, and the slow-change blocks are processed in the updating and positioning process.
  5. 5. The method for updating dynamic map construction and block differentiation according to claim 4, wherein the specific steps of the ground extraction algorithm are as follows: A1, converting laser point cloud into polar coordinates by taking the position of the mobile robot as the center, and partitioning according to the distance, angle and radius of the laser point cloud relative to the center to obtain point cloud partitions; A2, regarding each point cloud partition, taking a group of points with the lowest position as candidate ground points, fitting a local plane by using SVD, calculating the distance between the rest points in the point cloud partition and the local plane, and dividing the corresponding points into ground points in response to the distance being smaller than a preset threshold value to obtain the ground division point cloud extracted by the current frame; A3, converting the ground partition point cloud extracted from the previous frame into a current position coordinate system, supplementing the missing ground points of the current frame through multi-frame fusion, and carrying out weighted average on the existing ground points of the current frame to obtain a ground partition result.
  6. 6. The method for dynamic map construction and block differentiation updating as described in claim 1, wherein said step S3 comprises the sub-steps of: s31, performing ground separation on the original laser point cloud to obtain a ground point cloud and a non-ground point cloud; s32, directly storing the ground point cloud into the ground block, and simultaneously dividing the non-ground point cloud into blocks according to space; S33, uniformly setting the initial types of all map blocks in the map generated for the first time as slow change blocks.
  7. 7. The method for dynamic map construction and block differentiation updating according to claim 1, wherein the formula for assigning different weights to different types of map blocks in step S4 is: Wherein the method comprises the steps of Representing the ith map piece, Representing the weight of the ith map piece, The table stabilizes the block weights and, Representing a slowly varying partition weight of the block, Representing low dynamic blocking weights, and having 。
  8. 8. The method for dynamic map construction and block differentiation updating according to claim 7, wherein the overall pose optimization objective function of the positioning and map registration in step S4 is: Wherein the method comprises the steps of Representing the map and observed laser error function, Map block representing cloud of current observation point A subset of the coverage area.
  9. 9. The method for dynamic map construction and block differentiation updating as described in claim 8, wherein said step S5 comprises the sub-steps of: S51, according to the map blocks with completed registration And the current observation point is clouded in the map block Subset of coverage areas Calculating a map block Rate of change of (2) : Wherein the method comprises the steps of Representing the number of points or voxel occupancy; s52, for the same map block, recording multiple change rates in a time window T, and calculating the average value of the multiple change rates Sum of variances ; S53, if Updating the map block type into stable blocks, if Updating the map block type to a low dynamic partition, and otherwise updating the map block type to a slowly varying partition, wherein Representing the empirical threshold of the stable block mean, Representing an empirical threshold of stable block variance, Representing a low dynamic block mean empirical threshold, Representing a low dynamic block variance empirical threshold.
  10. 10. The method for dynamic map construction and block differentiation updating according to claim 9, wherein said step S6 is specifically: For stable partitioning, default is not updated, and only the change rate is counted; aiming at slow change blocking or ground blocking, performing voxel downsampling on old map points to enable the old map points to be sparse, and fusing new observation point clouds: Wherein Representing voxel downsampling; for low dynamic partitioning, the latest observation point cloud replacement is directly used: 。

Description

Dynamic map construction and block differentiation updating method Technical Field The invention belongs to the technical field of SLAM, and particularly relates to a design of a dynamic map construction and block differentiation updating method. Background Instant localization and mapping (SLAM, simultaneous Localization AND MAPPING) is one of the core technologies in the fields of mobile robots, unmanned aerial vehicles, etc. In a large-scale, long-term running scenario, the SLAM system needs to solve not only the positioning problem caused by map change, but also the reliability problem of long-term maintenance and dynamic update of the map. The existing SLAM system has the following general problems: (1) The dynamic environment adaptability is insufficient, namely objects (ground, buildings, trees, vehicles and the like) with different dynamic characteristics exist in a scene, the environment can change along with time, incorrect matching is introduced during positioning, positioning accuracy is reduced and even is incorrect, and in addition, certain SLAM systems only perform static/dynamic two classification and cannot model a change rule in fine granularity. (2) The positioning and mapping coupling is too strong, residual dynamic or semi-dynamic data in the map of the existing SLAM system can directly interfere with positioning optimization, and meanwhile, the map updating strategy is single, so that map drift or pollution is easy to cause. (3) The map scale is limited, the global point cloud map of the existing SLAM system is integrally loaded, the memory and storage pressure is high, and the ultra-large scale or infinitely-expanded scene is difficult to support. Therefore, how to realize an extensible, updatable and long-term maintainable dynamic map representation and updating mechanism while ensuring positioning robustness becomes an important research direction of the current SLAM technology. Disclosure of Invention The invention aims to solve the problems of reduced positioning precision and deteriorated map quality caused by long-term changes of dynamic objects and environments in the existing SLAM system, and provides a dynamic map construction and block differentiation updating method which improves the robustness of positioning registration, reduces the interference of the dynamic objects, realizes classification management and differentiation updating of map blocks, ensures the long-term reliability and accuracy of maps, and supports dynamic loading and efficient updating of infinite maps. The technical scheme of the invention is that the dynamic map construction and block differentiation updating method comprises the following steps: s1, dividing the global map into a plurality of map blocks according to a fixed space size, and eliminating a high-dynamic target in real time at the front end of the SLAM through a moving object detection method. S2, automatically classifying each map block into a stable block, a slow change block, a low dynamic block or a ground block. S3, initializing and generating a map according to the original laser point cloud. S4, positioning and map registration are carried out on the map generated by initialization, and different weights are given to different types of map blocks. S5, calculating the change rate of the map blocks which are registered, and dynamically adjusting the map block types according to the mean value and the variance of the change rate of a plurality of times in a time window. And S6, updating the stable blocks, the slow change blocks, the low dynamic blocks and the ground blocks by adopting different updating modes respectively, so as to realize differential updating of map blocks of different types. Further, each map block obtained by dividing in step S1 stores point cloud data, statistical features and dynamic state parameters independently. Further, the map blocks divided in step S1 support loading and unloading as needed to achieve local access. Further, the stable blocks in step S2 are not changed, the slow-change blocks are continuous and slow in change, the low-dynamic blocks exist in a certain time scale and are obvious in change, the ground blocks are obtained by a ground extraction algorithm, and the slow-change blocks are processed in updating and positioning. Further, the ground extraction algorithm comprises the following specific steps: a1, converting the laser point cloud into polar coordinates by taking the position of the mobile robot as the center, and partitioning according to the distance, the angle and the radius of the laser point cloud relative to the center to obtain a point cloud partition. A2, regarding each point cloud partition, taking a group of points with the lowest position as candidate ground points, fitting a local plane by using SVD, calculating the distance between the rest points in the point cloud partition and the local plane, and dividing the corresponding points into ground points in response to the d