Method and apparatus for urban road recognition based on laser point cloud, storage medium, and device

Patent metadata

Field	Value
Publication number	US-10430659-B2
Application number	US-201515750106-A
Country	US
Kind code	B2
Filing date	Dec 8, 2015
Priority date	Aug 4, 2015
Publication date	Oct 1, 2019
Grant date	Oct 1, 2019

Abstract

Embodiments of the present disclosure disclose a method and apparatus for urban road recognition based on a laser point cloud. The method comprises: constructing a corresponding road edge model according to the laser point cloud acquired by a laser sensor; determining a height of a mobile carrier provided with the laser sensor and constructing a corresponding road surface model based on the height and the laser point cloud; eliminating a road surface point cloud and a road edge point cloud in the laser point cloud according to the road edge model and the road surface model, segmenting a remaining laser point cloud using a point cloud segmentation algorithm, and recognizing an object corresponding to a segmenting result. By estimating the height of the mobile carrier according to the laser point cloud and constructing a corresponding road surface model using the height, the efficiency and accuracy of constructing the road surface model are improved, thereby improving the efficiency and accuracy of recognizing corresponding objects.

First claim

What is claimed is:

1. A method for urban road recognition based on a laser point cloud, comprising:
constructing a corresponding road edge model according to the laser point cloud acquired by a laser sensor;
determining a height of a mobile carrier provided with the laser sensor and constructing a corresponding road surface model based on the height and the laser point cloud;
eliminating a road surface point cloud and a road edge point cloud in the laser point cloud according to the road edge model and the road surface model, segmenting a remaining laser point cloud using a point cloud segmentation algorithm, and recognizing an object corresponding to a segmenting result,
wherein the method is performed by one or more processors.

2. The method according to claim 1 , wherein the determining a height of a mobile carrier provided with the laser sensor and constructing a corresponding road surface model based on the height and the laser point cloud comprises:
estimating the height of the mobile carrier using the laser point cloud close to the mobile carrier provided with the laser sensor, and with the height as an initial input threshold of a preset regression algorithm, constructing a corresponding road surface model based on the laser point cloud.

3. The method according to claim 2 , wherein the estimating the height of the mobile carrier using the laser point cloud close to the mobile carrier provided with the laser sensor comprises:
projecting the laser point cloud to a polar grid map with coordinates of the laser sensor as an origin; and
performing Ransac regression to projection grids corresponding to the laser point cloud close to the laser sensor to estimate the height of the laser sensor.

4. The method according to claim 2 , wherein the regression algorithm is a Gaussian process regression, and
with the height as an initial input threshold of a preset regression algorithm, constructing a corresponding road surface model based on the laser point cloud comprises:
performing Gaussian process regression to the projection grid corresponding to each frame of the laser point cloud according to the initial input threshold to obtain a corresponding candidate road surface point cloud; and
performing merging and spline regression processing to the candidate surface point cloud to obtain the road surface model.

5. The method according to claim 1 , wherein the constructing a corresponding road edge model according to the laser point cloud acquired by a laser sensor specifically comprises:
recognizing the laser point cloud using a corner point detection algorithm, to obtain road edge corner points corresponding to the laser point cloud; and
constructing the road edge model according to the obtained road edge corner points.

6. The method according to claim 1 , wherein the segmenting a remaining laser point cloud using a point cloud segmentation algorithm comprises:
clustering the remaining laser point cloud to obtain a corresponding laser point cloud cluster;
establishing a supervoxel corresponding to the laser point cloud cluster; and
segmenting the supervoxel to obtain laser point cloud sub-clusters, and performing merge processing to the laser point cloud sub-clusters.

7. The method according to claim 6 , wherein the establishing a supervoxel corresponding to the laser point cloud cluster specifically comprises:
establishing the supervoxel corresponding to the laser point cloud cluster based on spatial coordinates and reflectivity corresponding to the laser point cloud cluster.

8. The method according to claim 6 , wherein the performing merge processing to the laser point cloud sub-clusters specifically comprises:
obtaining shape characteristics of the laser point cloud sub-clusters through a principal component analysis; and
performing merge processing to the laser point cloud sub-clusters based on the obtained shape characteristics.

9. An apparatus for urban road recognition based on a laser point cloud, comprising:
a road edge modeling unit configured to construct a corresponding road edge model according to the laser point cloud acquired by a laser sensor;
a road surface model unit configured to determine a height of a mobile carrier provided with the laser sensor, and construct a corresponding road surface model based on the height and the laser point cloud;
a point cloud eliminating unit configured to eliminate a road surface point cloud and a road edge point cloud in the laser point cloud according to the road edge model and the road surface model;
a point cloud segmenting unit configured to segment a remaining laser point cloud using a point cloud segmentation algorithm; and
an object recognizing unit configured to recognize an object corresponding to a segmenting result.

10. The apparatus according to claim 9 , wherein the road surface model unit comprises:
a height estimating subunit configured to estimate the height of the mobile carrier using the laser point cloud close to the mobile carrier provided with the laser sensor; and
a road surface constructing subunit configured to, with the height as an initial input threshold of a preset regression algorithm, construct a corresponding road surface model based on the laser point cloud.

11. The apparatus according to claim 10 , wherein the height estimating subunit is specifically configured to:
project the laser point cloud to a polar grid map with coordinates of the laser sensor as the origin; and
perform Ransac regression to projection grids corresponding to the laser point cloud close to the laser sensor to estimate the height of the laser sensor.

12. The apparatus according to claim 10 , wherein the regression algorithm is a Gaussian process regression, the road surface constructing subunit is specifically configured to:
perform Gaussian process regression to the projection grid corresponding to each frame of the laser point cloud according to the initial input threshold to obtain a corresponding candidate road surface point cloud; and
perform merging and spline regression processing to the candidate surface point cloud to obtain the road surface model.

13. The apparatus according to claim 9 , wherein the road edge model unit comprises:
a corner point obtaining subunit configured to recognize the laser point cloud using a corner point detection algorithm to obtain road edge corner points corresponding to the laser point louds; and
a road edge constructing subunit configured to construct the road edge model according to the obtained road edge corner points.

14. The apparatus according to claim 9 , wherein the point cloud segmenting unit comprises:
a point cloud cluster unit configured to cluster the remaining laser point cloud to obtain a corresponding laser point cloud cluster;
a supervoxel subunit configured to establish a supervoxel corresponding to the laser point cloud cluster;
a sub-point cloud subunit configured to segment the supervoxel to obtain laser point cloud sub-clusters; and
a merge processing subunit configured to perform merge processing to the laser point cloud sub-clusters.

15. The apparatus according to claim 14 , wherein the supervoxel subunit is specifically configured to:
establish a supervoxel corresponding to the laser point cloud cluster based on spatial coordinates and reflexivity corresponding to the laser point cloud cluster.

16. The apparatus according to claim 14 , wherein the merge processing subunit is specifically configured to:
obtain shape characteristics of the laser point cloud sub-clusters through a principal component analysis; and
perform merge

Assignees

• Baidu online network technology beijing co ltd

Inventors

• Jiang Yu
• Yan Yang

Classifications

• G06V10/764

  using classification, e.g. of video objects · CPC title

• G06V20/39

  Urban scenes · CPC title

• G06V20/64 Primary

  Three-dimensional [3D] objects · CPC title

• G06F18/2132

  based on discrimination criteria, e.g. discriminant analysis · CPC title

• G06F18/2411

  based on the proximity to a decision surface, e.g. support vector machines · CPC title