Markov random fields for abnormal behavior detection on highways

摘要: This paper introduces a new paradigm for abnormal behavior detection relying on the integration of contextual information in Markov random fields. Contrary to traditional methods, proposed technique models local density object feature vector, therefore leading simple and elegant criterion classification. We develop Gaussian field mixture catering multi-modal integrating neighborhood into estimate. The convergence is ensured by online learning through stochastic clustering algorithm. system tested an extensive dataset (over 2800 vehicles) modeling. experimental results show that pedestrian walking, running cycling highway, detected with 82% accuracy at 10% false alarm rate, has overall 86% test data.