中文摘要：

为了对交通监控视频中的异常行为进行检测,需要对车辆的运动轨迹进行分析,但由于噪声、遮挡等原因,不可能获得完整的运动轨迹,导致分析结果不准确。针对此类问题,提出基于改进Hausdorff距离和谱聚类的轨迹聚类方法,首先对提取到的轨迹进行预处理,然后利用改进的Hausdorff距离进行轨迹相似度度量,最后通过谱聚类方法对距离矩阵进行聚类,得到符合实际情况的聚类结果。实验结果表明,该方法具有较好的鲁棒性和有效性。

英文摘要：

In order to detect abnormal behaviors from traffic monitor videos,it is necessary to analyze the trajectory of vehicles.However,because of noise,occlusion and other reasons,it is impossible to obtain the complete trajectory,thus causing the inaccuracy of analyzing results.To deal with problems like this,a trajectory clustering method based on improved Hausdorff distance and spectral clustering is proposed.First the extracted trajectories are pretreated;next the improved Hausdorff distance is used as trajectory similarity measurement;at last,by the spectral clustering method the distance matrix is clustered to acquire clustering results that are consistent with the actual situation.Experimental results show that the method bears fine robustness and effectiveness.