Method and device for processing a 3D point cloud representing surroundings

Patent metadata

Field	Value
Publication number	US-12190597-B2
Application number	US-202217744090-A
Country	US
Kind code	B2
Filing date	May 13, 2022
Priority date	May 18, 2021
Publication date	Jan 7, 2025
Grant date	Jan 7, 2025

Abstract

A method and to a device for processing a 3D point cloud representing surroundings, which is generated by a sensor. Initially, starting cells are identified based on ascertained starting ground points within the 3D point cloud which meet at least one predefined ground point criterion with respect to a reference plane divided into cells. Thereafter, cell planes are ascertained for the respective starting cells of the reference plane. Thereafter, estimated cell planes and ground points are ascertained for candidate cells deviating from the starting cells based on the cell planes of the starting cells, which are subsequently converted into final cell planes. As a result of such a cell growth originating from the starting cells, the cells of the reference plane are iteratively run through and processed so that the 3D point cloud is reliably classifiable into ground points and object points based on this method.

First claim

What is claimed is:

1. A method for processing a 3D point cloud representing surroundings, comprising the following steps:
receiving the 3D point cloud based on a sensor;
ascertaining starting ground points within the 3D point cloud, points of the 3D point cloud which are situated within a first predefined distance from the sensor and meet at least one predefined ground point criterion with respect to a reference plane being classified as the starting ground points, the reference plane being a plane which is predefined with respect to the sensor and being a plane which represents a ground surface;
dividing at least one sub-area of the reference plane into a plurality of cells, and ascertaining points of the 3D point cloud which correspond to each cell of the plurality of cells, points of the 3D point cloud corresponding to a respective cell of the plurality of cells being those points which are enclosed by the respective cell in a perpendicular projection onto the reference plane;
establishing cells of the plurality of cells which have a first predefined minimum number of the starting ground points as starting cells, the first predefined minimum number of the starting ground points including at least three of the starting ground points;
for each respective starting cell of the starting cells, ascertaining a cell plane, the cell plane being ascertained in such a way that it approximates a position of the starting ground points of the respective starting cell according to a predefined calculation rule;
ascertaining candidate cells within the reference plane, each candidate cell of the candidate cells:
being a cell for which no cell plane has been ascertained yet,
abutting at least one further cell for which a cell plane has already been ascertained, and
including a second predefined minimum number of corresponding points of the 3D point cloud;

calculating an estimated cell plane for each respective candidate cell of the candidate cells from all cell planes which are present in cells which directly abut the respective candidate cell;
ascertaining cell plane candidate points for each respective candidate cell of the candidate cells, each point of the 3D point cloud being ascertained as a cell plane candidate point for the respective candidate cell when:
the point corresponds to the respective candidate cell, and
a smallest distance from the point to the estimated cell plane of the respective candidate cell and/or from the point to the reference plane does not exceed a second predefined distance; and

for each respective candidate cell of the candidate cells, ascertaining a cell plane, the cell plane being ascertained in such a way that it approximates a position of the cell plane candidate points of the respective candidate cell according to the predefined calculation rule.

2. The method as recited in claim 1 , wherein:
(i) the plurality of cells which divide the reference plane abut one another without interruption, and/or are triangular or quadrangular cells, and/or
(ii) a shape and/or an extension of the plurality of cells is adapted as a function of a maximum resolution of the sensor, and/or existing boundary conditions.

3. The method as recited in claim 1 , wherein the at least one predefined ground point criterion is met when:
a shortest distance between respective points of the 3D point cloud and the reference plane does not exceed a third predefined distance, and/or
all vector products which result from possible combinations of vectors between a respective point of the 3D point cloud to be considered and at least two points directly adjoining the respective point of the 3D point cloud to be considered have a maximum permissible first angular deviation with respect to a normal of the reference plane, and/or
an average vector, ascertained from vector products which result from possible combinations of vectors between a respective point of the 3D point cloud to be considered and at least two directly adjoining points of the 3D point cloud, has a maximum permissible second angular deviation from the normal of the reference plane.

4. The method as recited in claim 1 , wherein points of the 3D point cloud are classified as object points when:
they are situated within the first predefined distance from the sensor,
their smallest distance from the reference plane or from a corresponding estimated cell plane or from a corresponding cell plane exceeds a predefined object minimum distance, and
their vector products with directly adjoining points exceed a third predefined angular deviation with respect to a normal of the reference plane.

5. The method as recited in claim 1 , wherein each respective point of the 3D point cloud, regardless of a prior classification as ground point, is classified as an object point when:
their closest adjoining point was classified as an object point,
a shortest distance from their closest adjoining point does not exceed a fourth predefined distance, and
a distance between the respective point and their closest adjoining point with respect to the reference plane does not exceed a fifth predefined distance.

6. The method as recited in claim 1 , wherein the predefined calculation rule effectuates a minimization of a sum of squared deviations between a respective cell plane and the corresponding starting ground points and/or cell plane candidate points of the corresponding starting cell and/or candidate cell for the respective cell plane.

7. The method as recited in claim 1 , wherein each cell plane of the cell planes ascertained for each starting cell of the starting cells and for each candidate cell of the candidate cells is only ascertained for a respective cell or used in a downstream processing step when:
a value for a distribution of starting ground points or cell plane candidate points within the respective cell reaches a predefined minimum distribution value, the value for the distribution being greater the further the points of the respective cell which have a largest distance from a center of the respective cell are situated away from the center of the respective cell, and the more uniformly the points of the respective cell are distributed within the respective cell, and/or
the cell plane does not exceed a maximum permissible angle of inclination with respect to the reference plane.

8. The method as recited in claim 1 , wherein initially an individual weighting of each cell plane of cell planes of adjoining cells takes place during the calculation of the estimated cell plane for each respective candidate cell of the candidate cells, the individual weighting being higher:
the higher a value is for a distribution of starting ground points and/or cell plane candidate points in a respective adjoining cell, and/or
the lower a sum is of squared deviations between a cell plane of a respective adjoining cell and corresponding starting ground points and/or cell plane candidate points in the respective adjoining cell.

9. The method as recited in claim 1 , wherein all points of the 3D point cloud within a cell including an ascertained cell plane which were not yet classified as ground points or as object points are classified as ground points when their smallest distance from the ascertained cell plane of the cell does not exceed the second predefined distance.

10. The method as recited in claim 1 , wherein a surroundings recognition is carried out in a surroundings recognition system based on ascertained cell planes and/or ascertained ground points and/or ascertained object points.

11. The method as recited in claim 1 , wherein the sensor is a LIDAR sensor of a transportation device.

12. A device for processing a 3D point c

Assignees

• Bosch Gmbh Robert

Inventors

• Fu Chengxuan
• Richter Jasmine
• Hardenacke Dennis
• Martins Costa Ricardo

Classifications

• G06T2207/20021

  Dividing image into blocks, subimages or windows · CPC title

• G06V20/588

  Recognition of the road, e.g. of lane markings; Recognition of the vehicle driving pattern in relation to the road · CPC title

• G01C7/02

  of land surfaces · CPC title

• G06T2207/10028

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

• G06T7/11

  Region-based segmentation · CPC title