Research On Improved Quantum Particle Swarm Optimization Based Hand Kinematics Tracking System

Since the Human-Computer Interaction(HCI)was introduced by Stuart K.Card,Allen Newell,and Thomas P.Moran et al.in the book Psychology of Human-Computer Interaction in the 1980 s,countries around the world are contributing their major force on the research of HCI.With the development of modern technology,the exploration in the complex scenario is becoming increasingly continual,which also complicates the tasks.How to make the computer understand human's intent through the most direct expression is becoming the main research on HCI.Among them,the human hand kinematics tracking based HCI system has great application prospects,which has been widely used in industry,national defense,space,ocean development,disaster relief,medical rehabilitation and other fields in many aspects of human life.In this paper,a model-fitting and optimization based tracking method was proposed to solve the problems in current human hand kinematics tracking system.In order to achieve the system,we first developed an improved quantum particle swarm optimization algorithm,which we call it DCQPSO,and designed a fully articulated simple human hand model.Finally,the human hand's kinematics information are obtained through the fitting of model hand and input hand optimized by DCQPSO.The three main parts of the paper is described as follow:Firstly,a fast,accurate and engineering oriented optimization algorithm is required.According to the mathematical analysis to the model of the tracking system,we adopt quantum particle swarm optimization algorithm as the core kernel of the system.Meanwhile,in order to alleviate the premature and convergence speed problem,two improved methods are proposed: Accelerate the convergence speed by modifying the attractor as a ? distribution random variable;the algorithm can obtain the real global optima during the search period with higher probability,by introducing a dynamic chaos mechanism,and the swarm is gifted a certain ability of escaping from local optima when the premature is detected.Indicated by the experiments on general benchmark function and variational autoencoder's inverse reconstruction,the improved methods are effective,and the DCQPSO outperforms both traditional particle swarm optimization and quantum particle swarm optimization.A further application to kinematics tracking is well grounded.Second,in order to obtain the precise hand joint angles,and kinematics,we established structural bionic hand skeleton model based on the basis of a full biological and mechanical analysis of the human hand.Mathematical relationship between each link joint coordinates is calculated through the Denavit Hartenberg parameters method,which maps the 26 Degree of Freedom(Do F)data space to the coordinate space,also called solution space.Then additional motion constraints are involved to the model to reduce the feasible solution space,and improve the optimization speed.Finally,a hand kinematics tracking system is achieved by optimization algorithm.We utilize the depth data of Kinect as the input data,and proposed a hand extraction preprocessing,then designed a rigorous and effective cost function of the DCQPSO,to accurately calculate the discrepancy of the input hand and model hand.The whole system can be seen under a “Model generating-Fitting” framework.Indicated by the experiments on human hand kinematics tracking in real world,this DCQPSO based tracking system can solve the complex hand kinematics tracking,and is robust.