Research in Knowledge-Based Vision Techniques for the Autonomous Land Vehicle Program
R. Nevatia, Keith Price, W. Otto Franzen, Salit Levy Gazit, Gérard Medioni · 1988
Abstract : This report describes our research in motion analysis and estimation techniques. This research is of particular relevance to the DARPA Autonomous Land Vehicle (ALV) program, but should also be of other general utility. Our basic approach detecting and tracking motion is to extract and match features, such as lines and regions, from a sequence and to generate emotion estimates from these. We present one report on matching edge elements in connected line segments (contours) in a sequence of views. This work assumes relatively small motions between views. We also present a report on an alternative representation for motion and a technique to use occlusion in spatio-temporal analysis. We also present results from a basic integrated system that combines feature extraction, matching and motion estimation. Keywords: Target detection knowledge-based vision.