基于编辑距离的轨迹相似性度量  

Trajectory Distance Metric Based on Edit Distance

作　　者：刘坤[1] 杨杰[1] 

机构地区：[1]上海交通大学图像处理与模式识别研究所,上海200240

出　　处：《上海交通大学学报》2009年第11期1725-1729,共5页Journal of Shanghai Jiaotong University

基　　金：国家高技术研究发展计划(863)项目(2007AA01Z164);国家自然科学基金项目(60675023)

摘　　要：针对运动轨迹分析问题,提出了2种改进的度量轨迹间相似性的方法.研究了编辑距离在应用于运动轨迹时存在的缺陷,改进了实数代价编辑距离算法,以对轨迹发展的影响来衡量插入和删除的代价;设计了一种分段计算方法,可以更好地表示轨迹运行的变化和持续,得到更具区分性的结果.与常见的编辑距离算法及其他几种流行的算法进行对比,证明了这2种算法在性能上的优势,也验证了新算法在聚类和异常运动检测上的作用.Two methods were proposed to measure A revised version of edit distance with real penalty the similarity between trajectories for motion analysis. was proposed to adapt edit distance for motion trajecto- ry analysis. Another novel segmented method was also put forward to describe the change and continuity in motion patterns. Then more discernable and intuitive measurement was achieved. The comparative experiment with other prevalent methods shows the superior discriminative properties of our methods. The applications on clustering and abnormal motion detection were also demonstrated.

关 键 词：运动轨迹 轨迹距离 编辑距离 聚类 异常检测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]