Abstract: Vision based dynamic hand gesture recognition technology is studied,skin-color based Gauss model and a revised optical flow tracing algorithm are combined,and real-time hand gesture tracking in complex background is realized with characteristics of rapidity,accuracy and good robustness.Hidden Markov model(HMM) is adopted as training recognition algorithm for the dynamic hand gesture recognizer.Considering characteristics of dynamic hand gesture,the best state chain determination algorithm and HMM parameter optimization algorithm are deduced by modifying the reestimation formula of HMM parameter optimization algorithm and adjusting the scale factors of the algorithm.At last,the dynamic hand gesture recognition algorithm is successfully used in a network based remote robot control system.