Visual control of vehicles using two-view geometry

Abstract

This paper addresses the problem of visual control for mobile robots with nonholonomic motion constraints. The vision system consists of a fixed camera mounted on the robot and no odometry or additional sensors are used. We consider the usual framework in which the target is defined by an image taken previously at the desired position. Then, the control law drives the robot from the initial position to the desired one by processing image information extracted from the current and target images. We present a new approach consisting in a switching control law based on the two-view geometry without scene constraints. Our main contribution is that two controllers are defined and combined in the switching control law. One is based on the epipolar geometry and the other on the homography model. Both models have well-known degenerate cases or particular situations in which the corresponding control fails when used alone. Nevertheless, the designed approach takes advantage of both models avoiding the drawbacks of each one and allowing a smooth motion of the robot. Experimental evaluation is presented to show the performance of the approach.