A Novel Trajectory Pattern Learning Method Based on Sequential Pattern Mining

Trajectory pattern learning is an important and meaningful issue for intelligent visual surveillance system. This paper puts forward a novel trajectory pattern learning method through sequential pattern mining. In our method, the flow vectors are firstly quantified by fuzzy C means clustering method; then a modified Prefixspan algorithm is applied to mine the sequential patterns from the trajectory sequences; finally, an approximate string matching method is adopted to detect whether a given trajectory is anomaly or not. The simulation experiments on different scenes demonstrate that our method is feasible and effective.