Dynamic trajectory indexing method for real time multiple object motion recognition

Abstract

this paper, we present a method for real time recognition of multiple object motion. In the framework, 3D motion trajectories of different objects are represented by the view invariant Indexing of General Shape (IGS) method to extract the high level description. In the motion recognition stage, we design a Time Order Sequence (TOS) strategy to recognize the interactions among the parallel motions of multiple objects. As the parallel trajectories are indexed by IGS in time series, the temporal intervals of trajectories can be extracted directly from the indexing. The indexing is updated each time on receiving a new frame. Thus, this method is effective for real time applications. The system is tested to represent and recognize human motions with multiple body parts interaction. Experiments are conducted to evaluate the effectiveness of this method. Copyright © 2012 ibai-publishing.