基于两步跟踪算法的海关车辆监控研究*

针对实时性和鲁棒性要求比较高的海关卡口车辆视频监控问题,提出了一种基于改进的均值漂移算法和粒子滤波算法的两步跟踪算法。对海关车辆监控的目标图像采用YCbCr颜色空间建立初始帧目标模型,利用改进后的均值漂移算法找出候选目标,在跟踪相似度小于设定的阈值时再利用改进后的粒子滤波算法进行后续的跟踪。通过实验分析,验证了提出的算法既能保证均值漂移算法跟踪的实时性,也能保证粒子滤波算法跟踪的鲁棒性,具有较好的应用价值。

Research on customs vehicle monitoring based on two-step targets tracking algorithm

The vehicle video monitoring at customs bayonet has a very high demand for its real-time and robustness. In response to this requirement,this paper proposed a two-step tracking algorithm based on improved Mean-Shift algorithm and particle filter. Firstly,established the initial frame image target model of customs vehicle monitoring by YCbCr color space. Secondly, identified the candidate targets by the improved Mean-Shift algorithm. Once the track similarity was less than the set threshold, then the follow-up tracking will be instead by improved particle filter algorithm. Finally, experiment analysis shows that the proposed algorithm can guarantee the real-time of Mean-Shift algorithm, but also can ensure the robust of particle filter tracking, so it is of good application value.