基于自适应分数阶Kalman滤波的船舶视觉跟踪

Ship Visual Tracking with Adaptive Fractional Kalman Filter

为提高Kalman滤波算法的准确性和鲁棒性,提出一种基于自适应分数阶系统的Kalman滤波算法,设计状态噪声协方差选择的自适应机制,推导其数学过程。将该算法应用到船舶视觉跟踪中,选取不同河流的CCTV(Closed Circnit Television)船舶监控视频(包括不同情况下的内河船舶运动监控),针对不同船舶大小、复杂光照、不同明暗度、多船会遇和多船追越等情况进行船舶视觉跟踪。对不同分数阶下的跟踪误差和准确度进行分析,并将分数阶Kalman滤波器与整数阶滤波器相比,说明其具有更宽的参数选择范围和更高的跟踪精度。研究结果表明:与分数Kalman滤波和Kalman滤波相比,该自适应分数Kalman算法具有更小的中心位置误差和更好的跟踪精度,能很好地避免跟踪器的漂移效应,具有较强的鲁棒性和准确性。

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.