Method, apparatus, electronic device and readable storage medium for point cloud data processing

Patent metadata

Field	Value
Publication number	US-11830207-B2
Application number	US-202117206999-A
Country	US
Kind code	B2
Filing date	Mar 19, 2021
Priority date	May 18, 2020
Publication date	Nov 28, 2023
Grant date	Nov 28, 2023

Abstract

A method, an electronic device and a readable storage medium for point cloud data processing, which may be used for autonomous driving, are disclosed. The feature vectors of respective points in the first point cloud data and second point cloud data are pre-learned, and thus the feature vectors of the first key points may be determined directly based on the learnt second feature vectors of respective first neighboring points of the respective first key points in the first point cloud data, and the feature vectors of the candidate key points may be determined directly based on the learnt third feature vectors of the respective second neighboring points of respective candidate key points in the second point cloud data corresponding to the first key points.

First claim

What is claimed is:

1. A method for point cloud data processing, comprising:
obtaining first feature vectors of respective points in first point cloud data, and determining one or more first key points in the respective points in the first point cloud data according to the first feature vectors of said respective points;
obtaining second key points in second point cloud data corresponding respectively to respective first key points in the one or more first key points, according to the one or more first key points and a preset first conversion parameter between the second point cloud data and the first point cloud data; wherein the first point cloud data and the second point cloud data are point cloud data of a same scenario obtained from different view angles;
determining a first preset number of first neighboring points of each of the first key points according to the first point cloud data, and obtaining a fourth feature vector of each of the first key points with a trained neural network feature extraction model according to a second feature vector of each of the first neighboring points of each of the first key points, relative coordinates and laser reflection intensity of the first neighboring points;
determining a second preset number of candidate key points of each of the second key points corresponding to each of the first key points according to a preset search radius and a preset grid size; determining a first preset number of second neighboring points of each of the candidate key points according to the second point cloud data, and obtaining a fifth feature vector of each of the candidate key points with the trained neural network feature extraction model according to a third feature vector of each of the second neighboring points of each of the candidate key points, relative coordinates and laser reflection intensity of the second neighboring points; and
determining a matching point registered by each of the first key points, according to the fourth feature vector of each of the first key points and the fifth feature vector of each of the candidate key points.

2. The method according to claim 1 , wherein before the obtaining a fourth feature vector of each of the first key points according to a second feature vector of each of the first neighboring points of each of the first key points, the method further comprises:
obtaining the second feature vector of each of the first neighboring points of each of the first key points according to the first point cloud data by using a pre-created feature model.

3. The method according to claim 1 , wherein before the obtaining a fifth feature vector of each of the candidate key points according to a third feature vector of each of the second neighboring points of each of the candidate key points, the method further comprises:
obtaining the fifth feature vector of each of the candidate key points according to the second point cloud data by using a pre-created feature model.

4. The method according to claim 1 , wherein the determining a second preset number of candidate key points of each of the second key points corresponding to each of the first key points according to a preset search radius and a preset grid size comprises:
determining a search space of each of the second key points according to the preset search radius by taking grid voxels closest to each of the second key points as a search center of each of the second key points, wherein grid voxels in the second point cloud data are determined according to the preset grid size; and
determining central points of the second preset number of grid voxels in the search space of each of the second key points, as the candidate key points of each of the second key points corresponding to each of the first key points.

5. The method according to claim 4 , wherein the determining a second preset number of candidate key points of each of the second key points corresponding to each of the first key points according to a preset search radius and a preset grid size further comprises:
determining a repeated search space according to the search space of each of the second key points; and
determining central points of the second preset number of grid voxels in the repeated search space as candidate key points for two or more second key points.

6. The method according to claim 1 , wherein the determining a second preset number of candidate key points of each of the second key points corresponding to each of the first key points according to a preset search radius and a preset grid size comprises:
determining a search space of each of the second key points according to the preset search radius by taking each of the second key points as a search center of each of the second key points;
determining the second preset number of grid voxels according to the preset grid size in the search space of each of the second key points; and
determining central points of the second preset number of grid voxels as the candidate key points of each of the second key points corresponding to each of the first key points.

7. The method according to claim 1 , wherein the determining a matching point registered by each of the first key points, according to the fourth feature vectors of each of the first key points and the fifth feature vectors of each of the candidate key points comprises:
obtaining a similarity between each of the candidate key points of each of the second key points and each of the first key points corresponding to each of the second key points, according to the fourth feature vector of each of the first key points and the fifth feature vector of each of the candidate key points; and
determining the matching point registered by each of the first key points according to the similarity between each of the candidate key points of each of the second key points and each of the first key points corresponding to each of the second key points and each of the candidate key points.

8. An electronic device, comprising:
at least one processor; and
a memory communicatively connected with the at least one processor; wherein,
the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform a method for point cloud data processing, which comprises:
obtaining first feature vectors of respective points in first point cloud data, and determining one or more first key points in the respective points in the first point cloud data according to the first feature vectors of said respective points;
obtaining second key points in second point cloud data corresponding respectively to respective first key points in the one or more first key points, according to the one or more first key points and a preset first conversion parameter between the second point cloud data and the first point cloud data; wherein the first point cloud data and the second point cloud data are point cloud data of a same scenario obtained from different view angles;
determining a first preset number of first neighboring points of each of the first key points with a trained neural network feature extraction model according to the first point cloud data, and obtaining a fourth feature vector of each of the first key points according to a second feature vector of each of the first neighboring points of each of the first key points, relative coordinates and laser reflection intensity of the first neighboring points;
determining a second preset number of candidate key points of each of the second key points corresponding to each of the first key points according to a preset search radius and a preset grid size; determining a first preset number of second neighboring points of each of the candidate ke

Assignees

• Beijing Baidu Netcom Sci & Tech Co Ltd
• Beijing Baidu Netcom Sci & Tech Co Ltd

Inventors

• Yu Li
• Lu Weixin
• Wan Guowei
• Peng Liang
• Song Shiyu

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06T7/33 Primary

  using feature-based methods · CPC title

• G06F18/22

  Matching criteria, e.g. proximity measures · CPC title

• G06T7/344

  involving models · CPC title