摘要
为了快速、准确地对视频序列中的行人进行检测和跟踪,提出了一种基于运动区域的Mean-shift行人检测和跟踪算法。采用改进的背景差分的三帧间差分法对运动区域进行提取,并通过离线训练的HOG分类器对行人进行识别。提出采用基于Mean-shift的轮廓跟踪算法对行人进行实时跟踪。实验结果表明,该算法能够将不包含行人信息的区域进行筛选排除,在不降低检测效果的情况下明显提高了行人检测及跟踪的效率,具有较强的鲁棒性。
To detect and track the pedestrian quickly and accurately in the video sequences, a Mean-shift pedestrian detection and tracking algorithm based on motion region is presented. Firstly, using an improved three inter-flame difference method of back ground subtraction to extract motion region, and identifying the pedestrian by the offline trained HOG classifier. Then, the con tour tracking algorithm based on mean-shift is presented to track the pedestrian in real time. Experimental result shows that the algorithm can filter and exclude the area which not contains pedestrian information, significantly improve the efficiency of pedes- trian detection tracking without reducing the effect of detection, and have a strong robustness.
出处
《计算机工程与设计》
CSCD
北大核心
2014年第1期223-227,共5页
Computer Engineering and Design
基金
重庆市科委自然科学基金计划基金项目(CSTC2010BB2399)
