中文摘要：

针对目前城市道路监控视频中存在的事件检测及报警不及时问题,提出了一种快速有效地检测车辆路线和位置违章的方法。该方法不依靠传统的车辆检测和分割方法,而是首先利用Harris角点检测的方法提取车辆角点,然后用金字塔Lucas-Kanande（L-K）对角点进行快速跟踪,同时对跟踪的角点实行谱聚类,利用Bhattacharyya距离对角点和聚类中心进行度量,并将其进行归一化,由预先设定的阈值可以判断车辆是否发生了异常行为。与传统车辆异常行为检测方法相比,该方法具有方便、快速的优点,鲁棒性较强。

英文摘要：

Aiming at the existing untimely problems in urban road monitoring video of the incident detection and alarm which is not on time,we propose a kind of quick and effective method for detect violation vehicle route and position.This method does not rely on the traditional approach for detecting and segmenting multiple vehicles.We use the Harris corner detection method to extract vehicle corner,then use the pyramid of Lucas-Kanande tracking to track the Harris corner fast,then implement spectral clustering for the tracking Harris corner.We use the Bhattacharyya distance to measure and normalize.The Harris corner and clustering center we can judge whether the abnormally vehicle behavior happeneds with a predetermined threshold.The experimental results show that the method we proposed is robust and effective.