Studies On The Hand Gesture Recognition System Based On Monocular Vision

As a new human-computer interaction technique,hand gesture recognition become more and more popular in our daily life.There are many kinds of human-computer interaction techniques,such as face recognition,voice recognition,character recognition and so on.Human hands are the most flexible organs in the human body.We can make interaction more convenient and strengthen its general performance if we use hands as a way to talk with computer.Moreover,most people which have different ages,race,and physical condition can use it easily.At present,many companies produced their own hand gesture recognition system and most of them are based on special wearable devices and special camera.They are expensive and can't afford by ordinary family.So if we can use a cheap camera to realize the hand gesture recognition will be fine.This paper has studied four key steps of gesture recognition and achieved the distinguish between four different types of hand gestures.In the aspect of image segmentation,we used a test method which were combined with skin color and edge contour to exclude the interference from the face area.Through the experiment,we proved that this method can get the gesture area effectively.In the aspect of hand tracking,we used the traditional Mean Shift algorithm for tracking.Through the test,we found that the algorithm will affected by other candidate area's skin color during the process of hand movement and lead to the tracking failure.This paper was proposed an adaptive tracking method which was combined with moving target detection and Mean Shift algorithm to solve this problem.In the aspect of gesture recognition,this paper used the KELM algorithm and make the method applied in the gesture recognition system.This paper use the KELM algorithm to compared with the traditional neural network and found that the KELM algorithm had many advantageous properties such as more rapid calculation,high generalization performance and high accuracy.