基于Kalman预测和点模式匹配的多目标跟踪

Multi-object tracking based on Kalman filtering prediction and point matching algorithm

针对复杂背景下多运动目标的跟踪方法不能有效解决遮挡和高速运动等问题,提出一种Kalman预测与点模式匹配相结合的多目标跟踪方法。利用Kalman滤波预测目标在下一帧图像中的位置,以此位置为中心确定目标搜索区域,然后以点模式匹配进行搜索区域和目标模板进行匹配,有效地解决目标的旋转和轻微的遮挡问题。为了提高匹配速度和实时性,在点模式匹配中利用Kalman滤波对目标旋转角度的预测与修正;同时为了保证跟踪的鲁棒性、连续性及准确性,对目标模板的更新采用置信度二级判决门限。实验表明该方法具有较好的实时性,并能够有效地解决遮挡等问题。

Since the muti-object tracking method can not effectively overcome the target covering and high-speed movement in complex background, a new multi-object tracking algorithm based on point matching algorithm and Kalman filtering predic-tion is proposed. The possible position of a moving object in the next frame image is predicted by Kalman filtering. The position is regarded as the search center to determine the object seach region. The matching of the object template and the candidate re-gions is carried out with point matching algorithm to solve the problems of rotation and slight covering of the target. In order to improve the matching speed and real-time performance, the rotation angle of the target is predicted and corrected in the process of point matching by using Kalman filtering. The secondary judgment threshold of confidence is adopted for object template up-date to ensure the tracking robustness, continuity, stability and accuracy. The experimental results show that this approach has better real-time performance and can solve the covering problem effectively.