Dynamic scene understanding using temporal association rules

摘要: The basic goal of scene understanding is to organize the video into sets events and find associated temporal dependencies. Such systems aim automatically interpret activities in scene, as well detect unusual that could be particular interest, such traffic violations unauthorized entry. objective this work, therefore, learn behaviors multi-agent actions interactions a semi-supervised manner. Using tracked object trajectories, we similar motion trajectories clusters using spectral clustering technique. This set depicts different paths/routes, i.e., distinct taking place at various locations scene. A mining algorithm used mine interval-based frequent patterns occurring pattern indicates are linked based on their relationship with other set, use Allen's logic describe these relations. resulting generate association rules, which convey semantic information contained Our overall rules govern dynamics perform anomaly detection. We apply proposed approach two publicly available complex datasets demonstrate considerable improvements over existing techniques. Uses technique event recognition dynamic scenesTemporal then generated from patterns. These help model sequence cycle.Spatio-temporal anomalies identified detected hierarchical