Feature extraction and tracking for scanning range sensors

A solution to the problem of simultaneously extracting and tracking a piecewise-linear range representation of a mobile robot's local environment is presented. The classical framework of the extended Kalman filter fits this problem extremely well and the algorithm presented is immune to vehicle motion and active sensor reorientation during the finite capture time of the range scan. The paper also tackles the problems of fast, non-iterative, initialisation of feature tracks, and feature track management over multiple scans. (C) 2000 Elsevier Science B.V. All rights reserved.