Research And Application Of 3D Motion Trajectory Representation And Matching Based On Segmentation

Human behavior analysis aims to identify and analyze the movement of the human body and try to understand the intention,which have a lot of application prospects in the fields of video surveillance,human-computer interaction and so on.Three-dimensional motion trajectory contains a lot of dynamic time series information,which can be used to describe a lot of complex human movements such as the posture of the hand,the movement of the body and others.Therefore,it is meaningful to establish the dynamic motion description and recognition using the 3D motion trajectory for human behavior analysis.However,it is not reliable to describe motion using raw trajectory data directly.Affected by some factors,the trajectory will be different in the position,direction,scale,which leads to the inconsistent on the spatial shape.Although geometrical features such as curvature and torsion can describe the local characteristics of the trajectory well,it cannot describe the spatial structure of the trajectory.Thus,it is useful to design a method that could extract the global and local characterization of trajectory effectively.Furthermore,due to the dynamic characteristics of motion,the track length and speed of two trajectories with same meaning are not different.Therefore,we need to align the trajectories too.In this paper,we mainly study the trajectory representation method which can describe the global structure of motion trajectory based on segmentation and study the corresponding matching scheme.Major work includes:(1)A trajectory segmentation method is proposed,which divides the trajectory into a sequence of sub-trajectory based on the motion movement.These sub-trajectories are belonging to the basic shapes.The basic shapes,which compose of straight lines,plane curves,and left or right helix,describe the movement of the three dimensions changes respectively,and can represent any trajectory in space.(2)To describe the structure of trajectory,this article employed a representation method based on trajectory segment.In this paper,an approximate value of the area integral invariant is used as the local feature,and the spatial distribution of the trajectory segment is taken as a global feature to distinguish the trajectory segment with obscured shape.(3)A DTW-based trajectory segmentation matching mechanism is designed for this trajectory representation.This matching mechanism matches the segments and achieves segment-level alignment.(4)Design and implement a virtual training system for TCM Bone-setting teaching.This paper applies the motion trajectory representation and matching strategy to the record,measurement and recognition for TCM Bone-setting Manipulation.Then these methods are extended to establish the evaluation mechanism.