Adaptive control of a video camera for the automatic detection and tracking of mobiles

The paper presents a system for the automatic tracking of mobile targets by means of computer vision. The system consists of a wrist with two degrees of freedom (yaw and pitch) that controls a video camera. First, a simplified kinematic model is proposed that relates the position of the target´s centroid in the image plane with the pan and tilt rotation angles. In order to permanently maintain the target within the field of vision of the video camera, a prediction algorithm is presented. The paper also revises several potential estimation and prediction techniques, like Kalman filtering, the Luenberger observer, the RMLS algorithm and the adaptive gradient-descent method, which have been applied to predict the centroid position