Context-Aware Distance for Anomalous Human Trajectories Detection

摘要

In this paper, a novel methodology for the representation and distance measurement of trajectories is introduced in order to perform outliers detection tasks. First, a features extraction procedure based on the linear segmentation of trajectories is presented. Next, a configurable context-aware distance is defined. Our representation and distance are significant in that they weigh the relative importance of several relevant features of the trajectories. A clustering method is applied based on the distances matrix and the outliers detection task is performed in any of the clusters. The results of the experiments show the good performance of the method when applied in two different real data sets.