Stabilized active camera tracking system

We present a robust and real-time stabilized active camera tracking system (ACTS), which consists of three algorithmic modules: visual tracking, pan-tilt control, and digital video stabilization. We propose an efficient correlation-based framework for visual tracking module that is designed to handle the problems which severely deteriorate the performance of a traditional tracker. The problems that it handles are template drift, noise, object fading (obscuration), background clutter, intermittent occlusion, varying illumination in the scene, high computational complexity, and varying shape, scale, and velocity of the manoeuvring object during its motion. The pan-tilt control module is a predictive open-loop car-following control strategy, which moves the camera efficiently and smoothly so that the target being tracked is always at the center of the video frame. Video stabilization module is required to eliminate the vibration in the video, when the system is mounted on a vibratory platform such as truck, helicopter, ship, etc. We present a very efficient video stabilization method that adds no extra computational overhead to the overall system. It exploits the coordinates of the target, computed by the tracker module, to sense the amount of vibration and then filters it out of the video. The proposed system works at full frame rate (30 fps), and has been successfully used in uncontrolled real-world environment. Experimental results show the efficiency, precision, and robustness of the proposed stabilized ACTS.