基于联合多特征直方图的Mean Shift行人跟踪方法研究

Research on Pedestrian Tracking Method Based on Joint Multi-feature Histogram and Mean Shift

针对单一颜色特征和目标间遮挡导致跟踪性能差的缺陷,提出一种联合多特征直方图和Mean Shift算法相结合的行人跟踪方法。将目标颜色和边缘特征使用直方图模型进行描述,利用运动信息修正颜色、边缘直方图核函数,以降低算法受目标形变、背景干扰和局部遮挡的影响。融合多特征直方图信息构建目标和候选目标模型,将其嵌入到Mean Shift跟踪框架中,实现行人跟踪。针对目标遮挡丢失问题,提出四步搜索策略法,通过目标周围环境、运动等信息捕获丢失目标。实验结果表明,该方法具有较高的准确性,能实时有效地跟踪行人目标,检测到的速度、加速度等交通信息与实际采集的相匹配。

In view of the poor performance of traditional tracking algorithm caused by single color feature and target occlusion,a new pedestrian tracking method based on the joint multi-features histogram and mean shift was proposed.The color and edge features of a target were described by the histogram model,while the kernel function of the color and edge feature was modified by motion information to reduce the influence of target distortion,background interference and partial occlusion on the algorithm.The information of the multi-feature histogram was fused to establish the target model and candidate target model,which was embedded into the Mean Shift tracking framework to achieve pedestrian tracking.Finally,the four-step search strategy which can relocate the missing target according to the surrounding environment and motion information was put forward to overcome the target loss problem caused by blocking.Experimental results indicated that the proposed method can accurately track target in real-time.The detected traffic information such as velocity and acceleration matched the actual information.