Localization method and apparatus, electronic device, and storage medium

Patent metadata

Field	Value
Publication number	US-12430797-B2
Application number	US-202218566500-A
Country	US
Kind code	B2
Filing date	May 27, 2022
Priority date	Jun 1, 2021
Publication date	Sep 30, 2025
Grant date	Sep 30, 2025

Abstract

The present disclosure relates to a localization method and apparatus, an electronic device, and a storage medium. The localization method comprises: acquiring a target local semantic point cloud map, a global semantic grid map, and a laser point cloud map; performing pose identification on the basis of the target local semantic point cloud map and the global semantic grid map to obtain a set of candidate poses; and determining a target pose according to the laser point cloud map and the set of candidate poses.

First claim

What is claimed is:

1. A localization method, comprising:
acquiring a target local semantic point cloud map at a current position, global semantic grid maps at two or more levels, and a laser point cloud map, wherein the global semantic grid maps at different levels have different resolutions, wherein each grid of the global semantic grid map records a value of probability that a target semantic object exists in the grid;
performing pose identification on the basis of the target local semantic point cloud map and the global semantic grid map to obtain a set of candidate poses; and
determining a target pose according to the laser point cloud map and the set of candidate poses, wherein the target pose comprises a loop closure pose and/or a relocalization pose,
wherein, the performing pose identification on the basis of the target local semantic point cloud map and the global semantic grid map to obtain a set of candidate poses, comprises:
acquiring a target grid stack comprising grids with a probability value of greater than zero in the global semantic grid map at the highest level, wherein the level of each global semantic grid map is negatively correlated with the resolution;
calculating, based on the target local semantic point cloud map, a first evaluation index for each grid in the target grid stack and a first evaluation index value for at least one grid in another level of global semantic grid map, and determining, based on the first evaluation index value, a candidate pose to obtain a set of candidate poses,
wherein, the calculating, based on the target local semantic point cloud map, a first evaluation index for each grid in the target grid stack and a first evaluation index value for at least one grid in another level of global semantic grid map, and determining, based on the first evaluation index value, a candidate pose to obtain a set of candidate poses, comprises:
constructing a current map identical to the target local semantic point cloud map;
creating a current grid stack identical to the target grid stack, and controlling a current grid located at the top of the current grid stack to pop;
calculating a first evaluation index value for the current grid based on the current map; and
determining the candidate pose based on the grid for which the first evaluation index value is greater than a first index threshold.

2. The method of claim 1 ,
wherein, the determining the candidate pose based on the grid for which the first evaluation index value is greater than a first index threshold, comprises:
if the first evaluation index value is greater than a first index threshold, judging whether the target global semantic grid map in which the current grid is located is a semantic grid map at the lowest level or not, and if the target global semantic grid map in which the current grid is located is the semantic grid map at the lowest level, adding a target angle and corresponding position coordinates of the current grid in the global semantic point cloud map into the set of candidate poses as a candidate pose, and if the target global semantic grid map in which the current grid is located is not the semantic grid map at the lowest level, adding a corresponding grid of the current grid in the grid map at the next level into the current grid stack; returning to perform the operation of controlling the current grid located at the top of the current grid stack to pop, until the current grid stack is empty; wherein the target angle is an angle of rotation of the current map in relation to the target local semantic point cloud map; and
if the first evaluation index value is less than or equal to the first index threshold, returning to perform an operation of controlling the current grid located at the top of the current grid stack to pop, until the current grid stack is empty.

3. The method of claim 2 , further comprising:
when the current grid stack is empty, rotating the current map by a preset rotation angle, obtaining a rotated local semantic point cloud map as the current map, and returning to perform the operation of creating a current grid stack identical to the target grid stack, until the current map is rotated by one circle.

4. The method of claim 1 , wherein, the calculating a first evaluation index value for the current grid based on the current map, comprises:
translating the current map to move a corresponding coordinate point of the current grid in the current map to an origin position, to obtain a translated map;
acquiring a probability value that is recorded in a grid in the target global semantic grid map corresponding to each semantic feature point within the translated map, as a probability value of the semantic feature point; and
calculating the first evaluation index value for the current grid based on the probability value of each semantic feature point.

5. The method of claim 4 , wherein, the calculating the first evaluation index value for the current grid based on the probability value of each semantic feature point, comprises:
when the target global semantic grid map is not a semantic grid map at the lowest level, calculating a probability average for a plurality of semantic feature points based on the probability value of each semantic feature point, as the first evaluation index value for the current grid; and
when the target global semantic grid map is the semantic grid map at the lowest level, calculating a probability average of the plurality of semantic feature points and a penalty function average based on the probability value of each semantic feature point, and taking the difference between the probability average and the penalty function average as the first evaluation index value for the current grid, wherein the penalty function average is used to characterize the degree of mismatch between the current map and the target global semantic grid map.

6. The method of claim 1 , wherein, the determining a target pose according to the laser point cloud map and the set of candidate poses, comprises:
based on a corresponding first evaluation index for each candidate pose in the set of candidate poses, calculating a weighted average of each pose parameter of a plurality of candidate poses as an average pose parameter,
determining a search space by taking the average pose parameter as a center and discretizing the search space to obtain a plurality of to-be-searched poses;
employing an iterative nearest-point algorithm to determine information about matching between each to-be-searched pose and the laser point cloud map, and calculating a second evaluation index value for each to-be-searched pose based on the matching information; and
selecting a to-be-searched pose with the largest second evaluation index value as a target pose.

7. The method of claim 6 , before selecting a to-be-searched pose with the largest second evaluation index value as a target pose, further comprising:
determining that a standard deviation of target pose parameters of the to-be-searched poses for which the second evaluation index value is greater than a second index threshold is less than a corresponding preset standard deviation threshold.

8. The method according to claim 1 , before the acquiring a target local semantic point cloud map at the current position, global semantic grid maps at two or more levels, and a laser point cloud map, further comprising
generating the global semantic grid maps at two or more levels based on a global semantic point cloud map.

9. The method of claim 8 , wherein, the generating the global semantic grid maps at two or more levels based on a global semantic point cloud map, comprises:
performing grid division on the global semantic point cloud map according to a preset size, and

Assignees

• Beijing Youzhuju Network Tech Co Ltd

Inventors

• Zhu Minzhao
• Kong Tao
• Li Lei

Classifications

• G01C21/005

  with correlation of navigation data from several sources, e.g. map or contour matching (G01C21/30 takes precedence) · CPC title

• G06T2207/10028

  Range image; Depth image; 3D point clouds · CPC title

• G06T7/70 Primary

  Determining position or orientation of objects or cameras (camera calibration G06T7/80) · CPC title

• G06F18/23

  Clustering techniques · CPC title

• G06T17/05

  Geographic models · CPC title