摘要
从序列模式挖掘的角度对视频目标运动轨迹的分析和应用问题进行了研究,提出了一种基于改进PrefixSpan的频繁轨迹模式挖掘算法,并给出了基于所挖掘的频繁模式进行在线目标运动异常检测的方法。该方法对目标的运动轨迹进行量化编码,采用改进的PrefixSpan算法挖掘其中连续出现的频繁模式,通过字符串近似匹配的方法来检测当前运动轨迹所表示的目标行为是否异常。由于不需要计算两两轨迹之间的相似性,该方法可以应用于规模较大、分布模式数目难以确定场合下的视频目标轨迹分析问题。对仿真和真实场景的实验验证了该方法的有效性。
A modified PrefixSpan algorithm is put forward to analyze the video target＇s movement through their trajectories in this paper.In the algorithm, the trajectories are recoded through vector quantization.The modified PrefixSpan algorithm is utilized to mine the frequent and continuous patterns from them.An approximate string matching method is given to detect whether the video target＇s movement is abnormal or not.Since this method needn＇t to measure similarity among the trajectoties,it can be well used in occasion where the trajectory set scale is very large or the num of trajectory distribution pattern is difficult to determine.The experiments on the trajectories of different scenes show that the method is effective.
出处
《计算机工程与应用》
CSCD
北大核心
2011年第32期7-10,58,共5页
Computer Engineering and Applications
基金
中央高校基本科研业务费专项资金资助(No.10QG21)
