Disturbance observer based controller in object tracking

Object tracking is large and important area in the field of computer vision. In this paper the pan/tilt camera is used to track objects. The paper focuses on the control algorithms for the camera which are chosen and tuned to yield fast transients without overshooting. The image processing algorithms are realized using open computer vision library. Two setups are presented in the paper. The first one shows the case when the pan/tilt system is placed on the ground. In other words, the goal is to keep the centroid of the moving object in the middle of the picture by the rotation of the camera. The second case demonstrates the performance of the system when the camera is mounted to a flying quadcopter. In this scenario, the noise reduction is a much bigger issue. The P controller with constrained rate of change of the output, in other words the speed of the camera rotation, expanded by the disturbance observer to guarantee the zero steady state error was used. This controller's performance benefits from the low pass filters used in the disturbance observer. The real experiments were performed with the sampling time of 60ms.