Image based 3D movement statistical pattern analysis

摘要

Movement pattern analysis is an effective approach to detect anomalies and behavior prediction. Existing methods depend on known scenes in which objects move along a predefined path. Moreover, most of these methods investigate 2D movement patterns. It is desirable to have automatic object movement pattern construction reflecting the knowledge of the scene. This paper proposes an automatic learning system for 3D movement patterns. The movement path of each object is considered to be a member of a cluster. In order to learn the movement patterns, the movement path is hierarchically clustered using spatial and temporal information and each movement pattern is then represented by a Gaussian distribution. Subsequently, behavior prediction is investigated using the extracted statistical movement pattern. Finally, the performance of the proposed algorithm is evaluated by simulations. Results indicate that the proposed method has a better performance when movement paths are not on a single plane.