A Simple Method for Eccentric Event Espial Using Mahalanobis Metric

摘要

The paper presents an approach, which detects eccentric events in real time surveillance video systems (e.g., escalators), based on optical flow analysis of multitude behavour followed by Mahalanobis and χ~2 metrics. The video frames are flagged as normal or eccentric established on the statistical classification of the distribution of Mahalanobis distances of the normalized spatiotemporal information of optical flow vectors. Those optical flow vectors are computed from the small blocks of the explicit region of successive frames namely Region of Interest Image (RII), which is discovered by RII Map (RIIM). The RUM is obtained from specific treatment of foreground segmentation of moving subjects. The method essentially has been tested against a single camera data-set.