Abstract: A motion-based detection approach of the video events in the traffic crossing scene is proposed to detect the abnormal motions by using the moving trajectories.Aiming at the given application background，this method needs not targets tracking or detection，and detects the abnormal events by using the motion characters directly.Firstly，after moving pixels abstracting，the method uses both the moving pixels’ aggregates and the moving vectors to decide the moving targets’ categories（vehicles or pedestrians）.Then based on this，the moving trajectories of the targets are analyzed，and the motions are judged which has the obvious different trajectories from the exist trajectories or disobey the traffic rules distinctly to be the abnormal motions.This method is experimented with some samples.The results show that this method can detect the abnormal moving events in the crossing.

Key words: abnormality detection, moving pixels abstract, moving vector calculate, moving trajectory analyze

摘要： 提出了一种基于运动特征的交通路口视频奇异事件的检测的方法。该方法针对具体应用背景，不需要进行单个目标的跟踪和检测，而是直接运动运动特征进行检测。方法首先提取运动像素，并通过运动像素集合和运动矢量的大小判断运动目标种类（车辆或行人）。然后，在此基础上进一步分析目标的运动轨迹，将新出现的和已经有运动轨迹明显不一致的运动，和明显不符合交通规则的运动（如车辆的逆行）判定为异常运动。该方法在一定样本基础上进行了实验，实验结果表明，该方法可以较好地检测交通路口的奇异性运动事件。

关键词: 奇异性检测, 运动像素提取, 运动矢量计算, 运动轨迹分析