Independent component analysis of optical flow for robot navigation

In this paper, we apply independent component analysis (ICA) to optical flow for the visual navigation of robots. Independent components of optical flow are used for motion cognition. In the medial superior temporal (MST) area of the brain, it is observed that independent components of optical flow are recognized for motion understanding. Adopting this hypothesis for the motion cognition of human beings, we introduce an algorithm for the visual navigation of an autonomous robot that captures an image sequence with a camera mounted on itself. Our algorithm separates optical flow into independent flow components by ICA and detects the feasible region for robot navigation on an image, assuming that optical flow is a linear combination of flows on a dominant plane and in obstacle areas. We present some experiments on corridor path estimation using a synthetic image sequence and a real image sequence obtained by a mobile robot, and evaluate the stability of our algorithm.