An extended fuzzy SOM for anomalous behaviour detection

Analysis of motion patterns is an effective approach for gaining better understanding of human behaviour. Many methods have been proposed to tackle this problem. However, unsupervised approaches have been widely accepted for clustering motion patterns, due to the fact that no previous knowledge of the scene is required. The fuzzy self-organizing map (fuzzy SOM) is an unsupervised method which has been previously used for classifying motion patterns. However, it suffers from high computational cost when a large number of output neurons is required, especially with complex scenes. In this paper, we propose a novel approach for dealing with the number of output neurons of fuzzy SOM in a complex scene. The performance of our approach shows better results compared with the normal approach, and without any major effect on the computational cost.