Research On Key Technology Of Hand Interaction Based On Multi-Visual Characteristics

Virtual helmet is usually used in the simulation during the engineering application.However,the scene of eyes may be disturbanced by the helmet.The people can't confirm his hand location and cannot interact effectively,then feasible hand modeling is necessary here,hand should be displayed accracily and stably in the virtual scene by analysising.As the hand tracking and identification is the key of simulation system,this paper proposed hand characteristics based on multi-visual features,the algorithm can complete the robust location tracking and achieve the goal of gestures classification by Pareto-Optimality.Hand tracking is the basis and core problem of vision interaction in virtual reality.Due to the poor performance resulted from the existing hand tracking methods in the instances of movement,scale changes,complex backgrounds and so on,this paper presents a new hand tracking algorithm under particle filtering tracking framework which adopts oriented gradient local binary pattern(OG_LBP)descriptor integrating texture and contour information.Furthermore,via introducing infrared depth information,the new tracking algorithm combine the observation information of current frame in the stage of particle sampling and updating by artificial bee colony algorithm,which can overcome the degeneracy problem in particle filtering and improves hand tracking precision by optimize the space search.From experiments,this paper's algorithm can achieve an accurate and robust tracking performance in complex backgrounds.For hand gesture recognition problem,establish gestrue images in several viewpoints based on the apparent method.In traditional apparent searching,the searching strategy is comparing the nearest neighbor seaching nodes of images layer by layer,which may lead to the low efficiency.While this paper proposes the algorithm of seaching via Pareto-Optimality to improve the situation.In the paper,20 kinds of gesture sampled database with three freedom degrees is established,through the K-NN and Pareto algorithm comparative experiments,the algorithm proposed by the paper is better at robust and real-time classification.