Research On Human Motion Tracking Analysis Based On Non-rigid Body Model

With the rapid development of artificial intelligence and virtual reality,human motion tracking technology attracts much attention as a frontier direction of image processing.However,because of problems like moving object detection,human modeling and matching and randomness of human motion,there is a great deal of uncertainty in recovering human motion from video sequences.Therefore,improving the accuracy and robustness of human motion tracking technology is a problem to be solved.In respect of above problems,the research results obtained are as follows:(1)Aiming at eliminating interference in images,this thesis adopts illumination compensation and H-CrCb skin color detection technology.The results show that the two technologies can realize real-time extraction of moving objects under illumination.(2)Since human body has the characteristic of nonrigid,this thesis introduces nonrigid model to simulate human body.It can change its own shape and size by adjusting the parameters.Compared with rigid model,nonrigid model has obvious advantages in the simulation of deformed limb.(3)In order to solve the problem of matching nonrigid model with actual human body,this thesis uses alternating direction method of multipliers.This algorithm can produce body's data points cloud,generate and splice curved slices with the help of OpenGL to complete the model of the human body.(4)To overcome the randomness of human movement in motion tracking,this thesis proposes the improved sparsity-based collaborative mode algorithm.This algorithm combines the advantages of generative and classification tracking algorithms.And the updating scheme is improved.The results show that the improved tracking algorithm performances good in robustness and accuracy.(5)According to the experimental results and the principle of the algorithms,this thesis analyzes various tracking algorithms from robustness,real time,accuracy and other aspects.The results show that the proposed tracking algorithm has the best overall performance among them,but the processing speed is to be improved.