Anomalous trajectory detection using the fusion of fuzzy rule and local regression analysis

Motion trajectories provide rich spatio-temporal information about an object activity. In this paper, we present a novel anomaly detection framework to detect anomalous motion trajectory using the fusion of local regression analysis and fuzzy rule-based method. That is, first of all we address the problem by segmenting our moving objects using a Gaussian mixture background model. Secondly, visual tracking using probabilistic appearance manifolds to extract spatio-temporal trajectory. Thirdly, local regression analysis and data quantization are performed on the extracted trajectory such that the anomalous detection can be performed as the incoming data are acquired. Finally, through the accumulative rank of local regression analysis and a fuzzy rule-based anomaly detection framework to detect the anomalous trajectory. Experimental results on various challenging trajectory data has validated the effectiveness of the proposed method.