Curvature Maxima-based Trajectories Mining
Shoji Hirano, Shusaku Tsumoto · 2010
In this paper, we present a method for trajectories mining that utilizes a multiscale comparison scheme based on curvature maxima. The method firstly identifies curvature maxima on a trajectory and traces their positions across scales in order to recognize the multiscale structure of the trajectory. Next, it searches for the structurally best matches between two input trajectories by comparing their sub trajectories in a cross-scale manner. After that, it calculates the value-based dissimilarity for each pair of the matched patrial trajectories and aggregates them into the final dissimilarity between the two trajectories. We evaluated this method on the UCI character trajectory dataset and on a real-world medical dataset. Experimental results showed that the method yielded good clustering results comparable to DTW and provided interesting clusters that might reflect the distribution of fibrotic stages.