Floor plane recovery from monocular vision for autonomous mobile robot on indoor environments

Abstract

Floor plane recovery is a very important issue on mobile robotics. Without the using of specialized hardware as Laser Range Finders (LRF) or some others optical time-of-flight sensors, plane recovery is really a challenge. In this paper we present a robust approach for plane recovery from monocular images coming from an uncalibrated active Pan/Tilt/Zoom (PTZ) camera mounted on a mobile robot platform. Floor plane is obtained by tracking features, as interest points, between two sequential images taken by the robot camera. Projective invariants are measured over groups of five non-three collinear feature points, in order to detect groups of points on the same plane. Then homographies are calculated for each invariant group of points and finally by a vote system the homography with a higher value, determines the major plane on the image sequence.