Method for automatically producing map data, and related apparatus

Patent metadata

Field	Value
Publication number	US-12196572-B2
Application number	US-202217961930-A
Country	US
Kind code	B2
Filing date	Oct 7, 2022
Priority date	Dec 23, 2021
Publication date	Jan 14, 2025
Grant date	Jan 14, 2025

Abstract

The present disclosure provides a method and apparatus for automatically producing map data. The method includes: performing track rectification on crowdsourcing tracks based on corresponding standard tracks, and locating each map element included, based on depth information of track point images included in the rectified crowdsourcing tracks; comparing a latest map element obtained based on the rectified crowdsourcing tracks locating and an old map element at a corresponding locating position using a pre-built entity semantic map; determining, in response to a change in the latest map element compared to the old map element, a target processing method according to a processing standard of a changed map element pre-abstracted from a map element update specification; and processing the latest map element according to the target processing method to obtain a processed latest map.

First claim

What is claimed is:

1. A method for automatically producing map data, the method comprising:
performing track rectification on crowdsourcing tracks based on corresponding standard tracks, and locating each map element in track point images included in the rectified crowdsourcing tracks, based on depth information of the track point images;
comparing a latest map element obtained based on the rectified crowdsourcing tracks locating and an old map element at a corresponding locating position using a pre-built entity semantic map;
determining, in response to a change in the latest map element compared to the old map element, a target processing method according to a processing standard of a changed map element pre-abstracted from a map element update specification; and
processing the latest map element according to the target processing method to obtain a processed latest map.

2. The method according to claim 1 , wherein the performing track rectification on crowdsourcing tracks based on corresponding standard tracks, and locating each map element in track point images included in the rectified crowdsourcing tracks, based on depth information of the track point images, comprises:
acquiring crowdsourcing tracks and standard tracks respectively collected for a same area, the crowdsourcing tracks and the standard tracks both comprising a plurality of track points, and each track point corresponding to at least one track point image;
rectifying the crowdsourcing tracks based on matched image features in corresponding track point images between the crowdsourcing tracks and the standard tracks, to obtain the rectified crowdsourcing tracks;
determining depth information of each map element in the track point images included in the rectified crowdsourcing tracks; and
locating each of the map element based on the depth information.

3. The method according to claim 2 , wherein the acquiring crowdsourcing tracks and standard tracks respectively collected for a same area, comprises:
acquiring crowdsourcing tracks of a target area collected by crowdsourcing users through personal devices, wherein the crowdsourcing tracks comprise a plurality of crowdsourcing track points, and each of the crowdsourcing track points corresponds to at least one crowdsourcing image; and
acquiring standard tracks of the target area collected by professional equipment, wherein the standard tracks comprise a plurality of standard track points, and each of the standard track points corresponds to at least one standard image.

4. The method according to claim 3 , wherein the rectifying the crowdsourcing tracks based on matched image features in corresponding track point images between the crowdsourcing tracks and the standard tracks, to obtain the rectified crowdsourcing tracks, comprises:
performing a preset single-track point matching operation on each of the crowdsourcing track points to obtain a track point matching degree between each of the crowdsourcing track points in the crowdsourcing tracks and each candidate track point;
wherein the single-track point matching operation comprises: determining a standard track point whose distance from the current crowdsourcing track point does not exceed a preset distance as a candidate track point; extracting respectively from a crowdsourcing image corresponding to the current crowdsourcing track point and a standard image corresponding to the candidate track point, to obtain a crowdsourcing image feature and a standard image feature; and determining the track point matching degree between the current crowdsourcing track point and the candidate track point, based on a feature matching degree between the crowdsourcing image feature and the standard image feature;
performing probability calibration on a location of each of the crowdsourcing track points based on the track matching degree, to obtain a calibrated position, a variance, and a probability of each of the crowdsourcing track points; and
fusing the variance as a confidence parameter of a probability model into Kalman filtering, and reconstructing each of the crowdsourcing track points based on a filtered position, the variance, and velocity to obtain the rectified crowdsourcing tracks.

5. The method according to claim 4 , wherein the extracting respectively from a crowdsourcing image corresponding to the current crowdsourcing track point and a standard image corresponding to the candidate track point, to obtain a crowdsourcing image feature and a standard image feature, comprises:
inputting the crowdsourcing image corresponding to the current crowdsourcing track point and the standard image corresponding to the candidate track point respectively into a pre-built model of local feature point matching based on attention mechanism; and
acquiring the crowdsourcing image feature and the standard image feature output by the model, respectively.

6. The method according to claim 4 , wherein the determining the track point matching degree between the current crowdsourcing track point and the candidate track point, based on a feature matching degree between the crowdsourcing image feature and the standard image feature, comprises:
calculating an initial feature matching degree between the crowdsourcing image feature and the standard image feature;
performing geometric verification on the crowdsourcing image feature and the standard image feature using a random sampling consensus algorithm, and using a number of inliers as a geometric verification result;
normalizing the geometric verification result to obtain a normalized geometric verification result having a same metric as the initial feature matching degree;
determining a final feature matching degree based on the initial feature matching degree and the normalized geometric verification result; and
determining the track point matching degree between the current crowdsourcing track point and the candidate track point based on the final feature matching degree.

7. The method according to claim 2 , wherein the determining depth information of each map element in the track point images included in the rectified crowdsourcing tracks, comprises:
inputting the track point images included in the rectified crowdsourcing tracks into a preset depth estimation model, wherein the depth estimation model is used to represent a corresponding relationship between a map image and depth information of a map element, training to obtain input samples in training samples of the depth estimation model as image data, and output samples as a depth map, and the depth map is obtained by registering point cloud data with the image data; and
acquiring the depth information of the corresponding map element output by the depth estimation model.

8. The method according to claim 1 , wherein the method further comprises:
identifying an attribute of each map element based on image identification, image classification and semantic segmentation technologies; and
determining a multi-dimensional semantic feature, based on an image feature of the map element obtained based on an image representation learning method and a scenario feature of the map obtained based on image scenario analysis, and constructing to obtain the entity semantic map based on the multi-dimensional semantic feature.

9. An electronic device, comprising:
at least one processor; and
a memory communicatively connected to the at least one processor;
wherein the memory stores instructions executable by the at least one processor, and the instructions, when executed by the at least one processor, cause the at least one processor to perform operations, the operations comprising:
performing track rectification on crowdsourcing tracks based on corresponding standard tracks, and lo

Assignees

• Beijing Baidu Netcom Sci & Tech Co Ltd

Inventors

• Xia Deguo
• Huang Jizhou
• Wang Haifeng

Classifications

• G01C21/387

  Organisation of map data, e.g. version management or database structures · CPC title

• G01C21/3848

  Data obtained from both position sensors and additional sensors · CPC title

• G01C21/3811 Primary

  Point data, e.g. Point of Interest [POI] · CPC title

• G01C21/3822

  Road feature data, e.g. slope data · CPC title

• G01C21/3841

  Data obtained from two or more sources, e.g. probe vehicles · CPC title