Ground plane estimation in a computer vision system

Patent metadata

Field	Value
Publication number	US-10890445-B2
Application number	US-201816185256-A
Country	US
Kind code	B2
Filing date	Nov 9, 2018
Priority date	Jan 5, 2016
Publication date	Jan 12, 2021
Grant date	Jan 12, 2021

Abstract

Estimation of the ground plane of a three dimensional (3D) point cloud based modifications to the random sample consensus (RANSAC) algorithm is provided. The modifications may include applying roll and pitch constraints to the selection of random planes in the 3D point cloud, using a cost function based on the number of inliers in the random plane and the number of 3D points below the random plane in the 3D point cloud, and computing a distance threshold for the 3D point cloud that is used in determining whether or not a 3D point in the 3D point cloud is an inlier of a random plane.

First claim

What is claimed is:

1. A method for ground plane estimation in a three dimensional (3D) point cloud in a computer vision system, the method comprising:
receiving a 3D point cloud generated based on a plurality of 2D frames captured by a monocular camera;
determining a distance threshold for the 3D point cloud based on an estimated height of a ground plane in the 3D point cloud;
estimating the ground plane of the 3D point cloud by performing the following for a number of iterations:
identifying a random plane in the 3D point cloud from three randomly selected non-collinear 3D points in the 3D point cloud, wherein an incline of the random plane meets pitch and roll constraints;
computing a cost function of the random plane, wherein the cost function is based on a difference between a number of inliers of the random plane and a number of 3D points below the random plane, wherein the distance threshold is used to determine whether or not a 3D point in the 3D point cloud is an inlier; and
saving the cost function as a best cost function if the cost function is better than a previously computed cost function for a previously identified random plane.

2. The method of claim 1 , further comprising pruning the 3D point cloud prior to estimating the ground plane to eliminate 3D points not likely to be on the ground plane.

3. The method of claim 2 , wherein pruning comprises sorting the 3D points in the 3D point cloud according to values of Y coordinates, and eliminating all 3D points from the 3D point cloud not included in a percentage of the sorted 3D points having the highest Y coordinate values.

4. The method of claim 3 , wherein the percentage is in a range of 45% to 55%.

5. The method of claim 1 , wherein computing a cost function comprises giving more weight to inliers of the random plane and 3D points below the plane that also lie in a trapezoid of a 2D frame of the plurality of 2D frames.

6. The method of claim 5 , wherein computing a cost function comprises computing the cost function as (a number of the inliers outside the trapezoid−a number of 3D points below the random plane outside the trapezoid)+w*(a number of the inliers inside the trapezoid−a number of 3D points below the plane inside the trapezoid), wherein w is a weight factor.

7. The method of claim 1 , wherein determining a distance threshold comprises determining a reference height for the ground plane based on 3D points in the 3D point cloud likely to be in the ground plane, and computing the distance threshold as a fraction of the reference height.

8. The method of claim 7 , wherein the fraction is based on the height of the monocular camera and a target distance threshold.

9. The method of claim 7 , wherein determining a reference height comprises sorting the 3D points in the 3D point cloud according to values of Y coordinates, and selecting a value of a Y coordinate of a 3D point at a percentile of the sorted 3D points as the reference height, wherein the percentile compensates for noise in the generation of the 3D point cloud.

10. A computer vision system comprising:
a monocular camera configured to capture a plurality of two dimensional (2D) frames of a scene; and
a processor configured to receive a three dimensional (3D) point cloud generated based on the plurality of 2D frames, the processor configured to:
determine a distance threshold for the 3D point cloud based on an estimated height of a ground plane in the 3D point cloud; and
estimate the ground plane of the 3D point cloud by performing the following for a number of iterations:
identifying a random plane in the 3D point cloud from three randomly selected non-collinear 3D points in the 3D point cloud, wherein an incline of the random plane meets pitch and roll constraints;
computing a cost function of the random plane, wherein the cost function is based on a difference between a number of inliers of the random plane and a number of 3D points below the random plane, wherein the distance threshold is used to determine whether or not a 3D point in the 3D point cloud is an inlier; and
saving the cost function as a best cost function if the cost function is better than a previously computed cost function for a previously identified random plane.

11. The computer vision system of claim 10 , wherein the processor is further configured to prune the 3D point cloud prior to estimating the ground plane to eliminate 3D points not likely to be on the ground plane.

12. The computer vision system of claim 11 , wherein the processor is configured to prune the 3D point cloud by sorting the 3D points in the 3D point cloud according to values of Y coordinates, and eliminating all 3D points from the 3D point cloud not included in a percentage of the sorted 3D points having the highest Y coordinate values.

13. The computer vision system of claim 12 , wherein the percentage is in a range of 45% to 55%.

14. The computer vision system of claim 10 , wherein computing the cost function comprises giving more weight to inliers of the random plane and 3D points below the plane that also lie in a trapezoid of a 2D frame of the plurality of 2D frames.

15. The computer vision system of claim 14 , wherein computing the cost function comprises computing the cost function as (a number of the inliers outside the trapezoid−a number of 3D points below the random plane outside the trapezoid)+w*(a number of the inliers inside the trapezoid−a number of 3D points below the plane inside the trapezoid), wherein w is a weight factor.

16. The computer vision system of claim 10 , wherein the processor is further configured to determine a distance threshold by determining a reference height for the ground plane based on 3D points in the 3D point cloud likely to be in the ground plane, and computing the distance threshold as a fraction of the reference height.

17. The computer vision system of claim 16 , wherein the fraction is based on the height of the monocular camera and a target distance threshold.

18. The computer vision system of claim 16 , wherein determining a reference height comprises sorting the 3D points in the 3D point cloud according to values of Y coordinates, and selecting a value of a Y coordinate of a 3D point at a percentile of the sorted 3D points as the reference height, wherein the percentile compensates for noise in the generation of the 3D point cloud.

Assignees

• Texas Instruments Inc

Inventors

• Nagori Soyeb
• Kumar Poorna
• Mathew Manu
• Viswanath Prashanth Ramanathpur
• Poddar Deepak Kumar

Classifications

• G06V20/56 Primary

  exterior to a vehicle by using sensors mounted on the vehicle · CPC title

• G01C3/08 Primary

  Use of electric radiation detectors · CPC title

• G06F18/2113

  by ranking or filtering the set of features, e.g. using a measure of variance or of feature cross-correlation · CPC title

• G06T7/77

  using statistical methods · CPC title

• G06T2207/30244

  Camera pose · CPC title