Adaptive Road Following using Self-Supervised Learning and Reverse Optical Flow

David Lieb and Andrew Lookingbill and Sebastian Thrun

Back to index

Summary

Assume vehicle is on the road. Follow regions of the same pattern. "Adaptive, self-supervised" learning algorithm using optical flow. Basically is alternate application for German Eichberger's "binning" approach that we tried. Data generated from actual DGC05 course. Not greatly applicable to DUC07, but interesting to see what Sebastian Thrun was up to prior to the DGC06.

Methods

Calculate image optical flow. Apply reverse optical flow, which uses previous images, to compute horizontal cross section templates. Match templates on existing image. Apply dynamic programming to find the optimal horizontal position of the road at the remaining heights. Interpolate to provide road segmentation.

Keywords

optical flow, reverse optical flow, self-supervised learning.

Rating

8

Bibtex Entry

@INPROCEEDINGS{ Lieb -RSS-05,

AUTHOR = {David Lieb and Andrew Lookingbill and Sebastian Thrun},

TITLE = {Adaptive Road Following using Self-Supervised Learning and Reverse Optical Flow},

BOOKTITLE = {Proceedings of Robotics: Science and Systems},

YEAR = {2005},

ADDRESS = {Cambridge, USA},

MONTH = {June},

URL = {http://www.roboticsproceedings.org/rss01/p36.html}

}

Back to index