Visual Servoing for Robot Navigation

This paper presents an integrated visual servoing system for robot navigation. This system is able to pursue a moving object by controlling a camera mounted on a pan/tilt head, so that the moving object is maintained in the center of the image. The visual system has four capabilities: the target detection, the target motion model online identification, camera control for target tracking and target position estimation. In order to minimize the time required for the image target detection, the target is made of elementary features: colored circular object. The target detection consists of two stages algorithm: (i) a color classification stage, and (ii) a knowledge-based shape detection stage. The color classification stage utilizes the distribution of the target color in the HSV color space in order to obtain an initial set of candidate regions. The second stage of the detection scheme uses mathematical morphology operators for circular object detection. The camera control exploits the detection in conjunction with an affine fit between 2 consecutive images. After the affine fit has been made, the camera control parameters are estimated. Due to the fact that the perspective projection is a many to one mapping, we designed the servomotorcamera-target system as a time variant system. A two phases control strategy is implemented. During the first phase (initialization) the target dynamics is estimated. The second phase consists of a state feedback control strategy. The target depth is estimated by using the appearance similarity between the target and its image. This system runs continuously in time and updates the target localization at a frame-rate of 170μs on a Pentium 400 MHz PC.