Semantic Trajectory Based Video Event Detection

Video event detection is a key research problem for automatic video content understanding.Trajectories of object are the one important clue for video event detection.Currently,the most trajectory based event detection methods focus on the geometric characteristics of the trajectory and neglect the semantic information related to the trajectory.However,as we known the trajectory producing process is affected by the trajectory related semantic information,such as geographic information related to trajectory etc.Combine the semantic information related to the trajectory with raw trajectory information can make us obtain more knowledge of trajectory.Semantic trajectory provides a way to effectively combine trajectory information with semantic knowledge.In this paper,the semantic trajectory is applied into video event detection and proposed a semantic trajectory based method for video event detection.This method can transform the trajectory of the interested object in video to semantic trajectory and detect the video event based on the semantic trajectory.Moreover,this method provides an approach for depicting semantic characters of semantic trajectory and matching the trajectory with trajectory characteristic description.Finally,the authors demonstrate the effectiveness of the proposed method through the empirical studies.