Stereo Feature Tracking for visual odometry (document)

Created Date : 2011.2
Reference :
Robust and efficient stereo feature tracking for visual odometry
stereo odometry - a review of approaches
multiple view geometry

How to get 3D point when we know feature image points of right and left camera?
How to propagate error? if we get the 3D point that calculated including stereo distance error.
If we know translated two 3D point, How to optimize error of R, T?
This document introduces about these problems.

- contents -

① Stereo Image Point :
Left Image Image
Camera Parameters :
Focal Length f, Principal point , Baseline B
Homogeneous Point
Non-Homogeneous Coordinate
-(stereo odometry A Review of Approaches)
ing in Stereo Navigation L. Matthies
Noise Propagation
X point Gaussian
Mean , Covariance
X Point(3D) mean, covariance ?
f is nonlinear function of a random vector with mean , covariance
② 3D point Covariance
~ Multiple view Geometry Nonlinear Error Forward propagation
③ Estimation of motion parameters
3D points ,
X:before motion, i-th:interest point, Y:after motion
Unique solution
(X, Y will be disturbed by same amount of noise)
Mean square error
Becomes minimal?
Several solutions.
- A solution based a singular value decomposition.
- A solution based on Essential Matrix.
- A maximum likelihood solution.
④ Maximum likelihood solution

<Doc> <PDF>

If you have good idea or advanced opinion, please reply me. Thank you
(Please understand my bad english ability. If you point out my mistake, I would correct pleasurably. Thank you!!)


스테레오 카메라에서 특정 점에 대한 왼쪽 영상에서 x,y점 오른쪽 영상에서 x,y점 을 알때 3D point를 어떻게 구할까?
3D을 구했을때 스테레오 영상에서 포함된 에러가 3D point에 에러가 어떻게 전파될까?
이동된 두 3D point가 일을때 어떻게 하면 에러를 최소화하는 R, T를 구할수 있을까?
이런 질문들에 대한 내용에 대한 솔루션을 다룬다.

<Doc> <PDF>

좋은 의견이나 답변 남겨 주세요.

No comments:

Post a Comment