On-road stereo visual odometry without explicit pose determinations

Patent metadata

Field	Value
Publication number	US-9558424-B2
Application number	US-201514755324-A
Country	US
Kind code	B2
Filing date	Jun 30, 2015
Priority date	Jun 30, 2015
Publication date	Jan 31, 2017
Grant date	Jan 31, 2017

Abstract

A method determines motion between first and second coordinate systems by first extracting first and second sets of keypoints from first and second images acquired of a scene by a camera arranged on a moving object. First and second poses are determined from the first and second sets of keypoints. A score for each possible motion between the first and the second poses is determined using a scoring function and a pose-transition graph constructed from training data where each node in the post-transition graph represents a relative pose and each edge represents a motion between two consecutive relative poses. Then, based on the score, a best motion is selected as the motion between the first and second coordinate systems.

First claim

We claim:

1. A method for determining a motion between a first coordinate system and and a second coordinate system, comprising steps of:
extracting a first set of keypoints from a first image acquired of a scene by a camera arranged on a moving object;
extracting a second set of keypoints from a second image acquired of the scene by the camera;
determining first and second poses from the first and second sets of keypoints, respectively;
determining a score for each possible motion between the first and the second poses using a scoring function and a pose-transition graph constructed from training data where each node in the post-transition graph represents a relative pose and each edge represents a motion between two consecutive relative poses; and
selecting, based on the score, a best motion as the motion between the first and second coordinate systems, wherein the steps are performed in a processor.

2. The method of claim 1 , wherein the keypoints are obtained using Speeded Up Robust Features (SURF).

3. The method of claim 1 , wherein the pose-transition graph is constructed using training data obtained from a video sequence acquired by the camera.

4. The method of claim 1 , wherein the poses represented in the pose-transition graph are determined using an analytical motion estimation procedure.

5. The method of claim 4 , wherein the relative motions are obtained using analytical motion estimation and 5-point 2D to 2D motion estimation procedure.

6. The method of claim 4 , wherein the relative motions are obtained using analytical motion estimation and 3-point 2D to 3D motion estimation procedure.

7. The method of claim 1 , wherein the poses are determined using a geographical positioning system (GPS).

8. The method of claim 1 , wherein the poses are determined using an inertial measurement unit (IMU).

9. The method of claim 1 , wherein the poses are clustered to remove similar motions to construct a compact pose-transition graph.

10. The method of claim 1 , wherein the scoring function depends on a number of inliers obtained from matching the first set of keypoints and the second second set of keypoints.

11. The method of claim 1 , wherein the motion is used for the task of visual odometry.

12. The method of claim 1 , wherein the motion is used for the task of localization using a 3D model of the scene.

13. The method of claim 1 , wherein the moving object is a vehicle.

14. The method of claim 1 , wherein the moving object is a robot.

Assignees

• Mitsubishi Electric Res Laboratories Inc

Inventors

• Ramalingam Srikumar
• Lee Gim Hee

Classifications

• G06T7/246 Primary

  using feature-based methods, e.g. the tracking of corners or segments · CPC title

• G06T11/26

  Drawing of charts or graphs · CPC title

• G06T7/0075

  Physics · mapped topic

• G06T11/206

  Physics · mapped topic

• G06T7/2086

  Physics · mapped topic