Research On Human Gesture Recognition Algorithm And Application Based On Virtual Reality

In order to improve competitiveness and work efficiency,it is necessary to obtain the progress of employees' work in real time.These are realized by recording work progress more frequently.But people needs to spend a long interval to record working status and human resources consumption.It is not ideal for such a record management system.In order to record work schedules more frequently and to avoid manpower consumption,it is a need for a scheme that automatically records progress without human beings.In this paper,the information of human joint point data and color video data collected by Microsoft Kinect are used to realize the estimation of unattended monitoring and progress recording,which is implemented on the virtual reality interactive platform.The main algorithms involves the extraction of behavior features,the detection of interest points,and the gesture recognition algorithm.Firstly,in order to solve the problem of different lengths of motion when the employee's motion is recognized and low recognition rate caused by occlusion of the skeleton,the motion recognition system based on dynamic time warping and the recognition system based on local features are respectively adopted.Secondly,in order to improve the accuracy and robustness of the motion recognition,the paper fused and corrected the recognition results of the two motion recognition systems.Finally,the interactive application of the recognition system based on the decision fusion of Bayesian algorithm and data correction technology is implemented on the constructed virtual reality platform to complete the task of unattended monitoring and progress record estimation.The gesture recognition system proposed in this paper is simulated on the Matlabplatform using two common data sets and one self-acquisition data set.The results showed that this recognition system based on Bayesian algorithm and data correction technology of decision fusion could improve the accuracy of the orderly motion recognition to 90%.The system realized 10 Actual Assessments on the Unity Platform.The recognition rate of the motion recognition system of this virtual training can reach about 90% for most actions and the sensitivity reaches about 90%.The system reached the accuracy and sensitivity requirements.