Human Action Recognition Based On Dense Optical Flow

In public places,such as hospitals,schools,banks,shopping malls and stations,airports,video surveillance has become an indispensable means of monitoring and deterring crime,but these cameras and related storage equipment can only take set and store video data,do not have the ability to identify the behavior of human behavior,the human actions and their behavior in the video still need manually analyze and calibration,it is a heavy workload with low efficiency.In the monitoring application of human action recognition technology can effectively reduce the work of the staff strength and improve work efficiency.The parts of this paper is as follows:First,we preconded the video,then get moving target foreground use take two-layer background modeling method.Second,generate motion history image,the Zernike moment and Hu moment of the motion history image are used as geature vectors to input to support vector machine.Third,we use Gunnar Farneback algorithm to obtain dense optical flow basis on motion history image,then get the histograms of oriented gradients,histogram of optical flow and moving boundary histogram,the global feature Hu moments and Zernike moments are used to multi feature fusion.Through the analysis of experimental results,the accuracy rate of human motion recognition Which is based on the Hu moments and Zernike moments of motion history image is 81.1%,the accuracy rate of human motion recognition Which is based on the Hu moments and Zernike moments of the dense optical flow of motion history image is 95.0%,The method of human motion recognition in this paper shows faster and more accurate performance.