摘要
针对欧氏距离和Hausdorff距离等在描述目标运动轨迹差异性时度量不够准确的问题,提出一种基于归一化编辑距离和谱聚类的轨迹分布模式学习方法.首先对目标的运动轨迹进行矢量量化编码;然后采用归一化的编辑距离来度量轨迹编码序列之间的差异,得到归一化编辑距离矩阵;再通过该矩阵进行谱聚类来提取轨迹的分布模式;最后利用所提取的轨迹分布模式确定整条轨迹及其局部是否异常.通过仿真和真实场景的实验验证了该方法的有效性.
For the inaccuracy problem of using Euclidean and Hausdorff distances to measure the trajectories＇ difference, a motion trajectory learning approach is developed based on the normalized edit distance and spectral clustering algorithm. Firstly, the trajectories are recoded through vector quantization. Then, a normalized edit distance is adopted to measure the difference among the trajectories. After that, the spectral clustering algorithm is applied to obtain the trajectories＇ distribution patterns based on the pair-wise distance matrix. Finally the learned patterns are used to detect the local and global anomaly. Experiments on synthetic and real world data sets demonstrate the effectiveness of our proposed approach to trajectory analysis and anomaly detection.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2008年第6期753-758,共6页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(60472072)
教育部博士点基金(20040699034)
关键词
轨迹模式
归一化编辑距离
异常检测
trajectory pattern normalized edit distance anomaly detection
