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

• DOI: 10.15607/rss.2005.i.036
• Publication Date: 2016-01-03T02:36:08Z
• Authors: David Lieb; Andrew Lookingbill; Sebastian Thrun
• Keywords: Optical Flow; Road surface; Identification; Tracking (education); Template; Template matching

The majority of current image-based road following algorithms operate, at least in part, by assuming the presence structural or visual cues unique to roadway.As a result, these are poorly suited task tracking unstructured roads typical desert environments.In this paper, we propose algorithm that operates selfsupervised learning regime, allowing it adapt changing conditions while making no assumptions about general structure appearance surface.An application optical flow techniques, paired with one-dimensional template matching, allows identification regions camera image closely resemble learned recent past.The assumes vehicle lies on order form templates road's appearance.A dynamic programming variant is then applied optimize 1-D match results enforcing constraint maximum curvature expected.Algorithm output images, as well quantitative results, presented for three distinct types encountered actual driving video acquired California Mojave Desert.