Surveillance Scene Modeling via Tracklet Analysis

A surveillance scene modeling algorithm is proposed by tracklet analysis. A number of tracklets belonging to the same object are found through a bidirectional spatio-temporal continuity tracklet association method. The entry, exit, main path, and scene occluder of a scene model are obtained by analyzing the reason that object's track fall into tracklets. Experiments prove that the scene model can improve tracking result and detect abnormal object tracks.