A Learning Method of Detecting Anomalous Pedestrian

摘要

Abnormal behavior detecting is one of the hottest but most difficult subjects in Monitoring System. It is hard to define "abnormal" in different scenarios. In this paper firstly the classification of motion is conducted, and then conclusions are made under specific circumstances. In order to indicate a pedestrian's movements, a complex number notation based on centroid is proposed. And according to the different sorts of movements, a set of standard image contours are made. Different behavior matrices based on spatio-temporal are acquired through Hidden Markov Models (HMM). A Procrustes shape analysis method is presented in order to get the similarity degree of two contours. Finally Fuzzy Associative Memory (FAM) is proposed to infer behavior classification of a walker. Thus anomalous pedestrians can be detected in the given condition. FAM can detect irregularities and implement initiative analysis of body behavior.