一种基于卡尔曼滤波的压缩跟踪算法研究

Compressive tracking algorithm based on Kalman filter

根据道路交通监控视频的特点,采用压缩跟踪(CT)算法进行运动车辆的检测与跟踪。在摄像头变化较大、运动车辆尺度变化和背景变化等情况下,CT算法均具有很强的鲁棒性。但是当车辆被遮挡时,跟踪算法容易失效。为了解决这一问题,提出使用卡尔曼滤波对遮挡的车辆进行轨迹预测。卡尔曼滤波能根据CT算法跟踪目标的轨迹,有效地预测目标遮挡时的轨迹。实验结果表明,本算法不但可以较好地处理跟踪车辆尺寸变化的问题,在车辆丢失或被部分遮挡时,能准确而稳定地跟踪车辆,而且具有很好的实时性,满足了工程应用的需求。

We employ compressive tracking （CT） algorithm to detect and track motion cars based on the characteristics of traffic monitoring video. The algorithm has strong robustness for greater camera change, motion vehicles scale change, and background change. However, the algorithm is easy to fail when a vehicle is sheltered. We therefore present a Kalman filter based modified CT algorithm to predict the motion trail of a sheltered vehicle. Kalman filter can track the trail of a target with CT algorithm, and effectively predict the trail of a sheltered target. Experimental results show that the algorithm can not only better solve the issue of motion vehicles scale change - precisely and stably track a vehicle when it disappears or is partly sheltered -, but also has better real-time performance. It thus satisfies the requirements of engineering application.