Ship Visual Tracking with Adaptive Fractional Kalman Filter

The Kalman filter algorithm based on adaptive fractional order system is introduced. An adaptive mechanism for selecting state noise covariance is designed and the deduction process is explained. The new algorithm is tested with CCTV（Closed Circnit Television） monitoring videos from rivers. The tracking performance is examined under a variety of conditions. The high accuracy and robustness of the new algorithm are demonstrated for all kinds of situations: different ship sizes, complex lighting, dim light, multi ship encounter and multi ship overtaking. The tracking error and accuracy of different fractional order Kalman filter are analyzed. It is shown that, compared with integer order filter, fractional order Kalman filter has wider parameter selection range and higher tracking accuracy.