Cluster-based trajectory overall trend extraction
Ailin He, Liu Zhong, Dechao Zhou · 2017
This paper proposed a trajectory overall motion trend extraction method based on trajectory clustering, including three phases: partitioning phase, clustering phase and extracting phase. Firstly, trajectories are partitioned at characteristic points whose turn angle or accumulated turn angle exceeds the threshold value determined by minimum description length (MDL) principle. Secondly, an improved distance measure function, which can express the distance between line segments with direction better when compared with the one in TRACLUS, is used to acquire trajectory line segment clusters. Lastly, original trajectory segments are used to extract the overall trajectory motion trend, which can obtain more details of the trajectory motion trend. Experiment results demonstrate that our algorithm can correctly extract the overall trajectory motion trend from real trajectory data.