Rch On Vision-based Hand Gesture Recognition

With the development of computer technology, people-centered human-computer interaction has become the focus of research. As an interactive interface, gesture is simple, efficient and user-friendly. And vision-based gesture solution can provide a non-contact interaction interface, which can meet the new demands of human-computer interaction. However, application of vision-based gesture recognition has encountered numerous problems. It is still a challenge to make the gesture interface barrier-free.The main work of this dissertation is as follows:(1) In the light compensation part, for light situation has great impact on image quality, an improved light compensation strategy was proposed. First, the dissertation used a method based on a logarithmic transformation and exponential transformation to adjust the brightness of the image. Then a white balance method based on dynamic threshold was applied for secondary adjustment. Experimental results show that this method can correct the darker and highlighted areas of the image.(2) In the static gesture recognition part, as gesture segmentation based on skin color was not good enough, an algorithm of hand segmentation combining HSV and YCbCr color space was given. And a classifier for static gestures recognition was implemented based on Support Vector Machine. Experimental results show that the modified hand segmentation method can give better result and the Support Vector Machine algorithm can improve the recognition rate significantly comparing with the traditional template matching method.(3) In dynamic gesture recognition part, in order to improve stability of hand tracking process, a modified Tracking-Learning-Detection method was proposed. A Haar classifier was applied to correct the deviation of the tracking process. And a classifier based on Hidden Markov Model was trained to classify the dynamic hand gesture. Comparative results show that the improved tracking algorithm can effectively improve the stability of the tracking process and the Hidden Markov Model algorithm can provide a better recognition accuracy.