一种基于DTW改进的轨迹相似度算法

Improved DTW algorithm for trajectory similarity

针对传统的轨迹相似度计算方法无法区分现实采集到的轨迹中由噪声带来的差异和真实的不相似部分的问题,基于动态时间规整(dynamic time warping,简称DTW)算法,提出了一种改进的轨迹相似度的计算方法。并对最后的结果进行了归一化处理,便于人们直观理解,同时也可用于对多对轨迹之间的相似性进行排序,从而可以在数据挖掘的相关应用中得到有效利用,同时对计算过程也进行了优化。在现实采集到的数据上的测试表明这种方法对噪声和异常点是鲁棒的,对轨迹的采样频率等参数没有任何要求,而且可以适用于仅获得轨迹的部分片段的情况,并且在区分轨迹的相似和不相似部分方面较之前的方法准确度有了很大提升,即使轨迹的采样较为稀疏的前提下依然如此。

To deal with the problem that traditional trajectory similarity algorithms always can’t distinguish noise and dissimilarity in trajectories,this paper proposed a new method to calculate the similarity between trajectories based on dynamic time warping（DTW）algorithms and normalized the final result for easy comprehension and can be used to rank multiple pairs of trajectories conveniently in data mining applications.The algorithm progress was also optimized.The experiment conducted on dataset sampled from real life indicated that this method is robust to noise and abnormal points,and does not have any assumptions on the sampling rate,it also works well even we only got partial trajectories.It also shows that our method has great advance in distinguishing the similar and dissimilar part of the trajectories,it also works when the trajectories are sparse sampled.