Moving Target Detection In Video Surveillance,Tracking And Gesture Recognition Algorithm Research

The construction of "smart city" is to promote the new fulcrum and new power of social economy, the rapid development of science and technology in today's world, and the world Chinese standards of science and technology, intelligent monitoring technology is the construction of "smart city" a necessary step, including video monitoring, video image detection, video image processing technology is also followed by rapid with the development of communication technology in recent years, the rapid development of China, other new technology industry rapid development, with the implementation of 4G technology, everyone holds a mobile terminal equipped with Android or IOS operating system, the telecommunications industry more convenient service to us, the relationship between people is more closely and convenientThe first step of monitoring objects were detected. In this paper, the design of the system is using improved background subtraction algorithm and three frame difference method "and" the result of the operation as a feedback, selective updates the background. This method can not only reduce the dynamic factors of error detection, but also robust to deal with target motion uncertainty and obstacles occlusion, illumination changes, dynamic scenes of target detection problem.The second step, tracking monitoring object, using the idea of CAMSHIFT and particle filter algorithm can alter adaptive tracking window size, greatly reducing the number of particles involved in the iterative, small amount of calculation and under partial occlusion and deformation than the traditional CAMSHIFT algorithm has better tracking effect.The third step, of monitored object for action recognition, used human minimum circumscribed rectangle the width to height ratio of, key gesture template is built using the hybrid wavelet moment features, using hidden Markov model algorithm of the test sample categories. The method can effectively improve the accuracy of the classification of human behavior.The system is using VS2010 operating system and call opencv database functions to analysis the movement characteristics of target tracking, using C + + programming language, the language of the monitored object detection, tracking and action recognition system, the details of each module in the system need to use the, finally points out some problems existed in this system, and the prospect for future development and algorithm improvement.