Uncertainty Quantification of Lucas Kanade Feature Track and Application to Visual Odometry

Xue Iuan Wong, Manoranjan Majji; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2017, pp. 10-18

Abstract


An uncertainty quantification approach to estimate the errors incurred by the Kanade Lucas Tomasi (KLT) feature tracking algorithm is presented. The covariance analysis is based on the linearized sensitivity calculations of the KLT algorithm. Track uncertainty thus computed is utilized to quantify the errors associated with feature based relative pose estimation algorithms. This paper also show that the uncertainty analysis result can serve as a mean to measures reliability of feature correspondences. Proposed technique show that a large amount of outlier can be ejected effectively, and thus improve the efficiency of iterative method such as RANSAC.

Related Material


[pdf]
[bibtex]
@InProceedings{Wong_2017_CVPR_Workshops,
author = {Iuan Wong, Xue and Majji, Manoranjan},
title = {Uncertainty Quantification of Lucas Kanade Feature Track and Application to Visual Odometry},
booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {July},
year = {2017}
}