Research On Static Gesture Recognition System Based On Monocular Vision

With the development of computer and information science technology rapidly, human computer interaction has become an indispensable part of people’s daily life. Since intuitive and natural is the characteristics of the gesture, gesture recogniton is an important branch of human computer interaction, as a key technology, gesture recognition has become a hot research topic in the field of image processing, pattern recognition, computer vision and so on. As a means of human-computer interaction, gesture is widely used in the fields of intelligent system, robot control, multimedia teaching, entertainment games, and communication between the deaf and dumb people.In this dissertation we design and implement a static gesture recognition system based on single camera. Be able to identify 26 English letters. Related to field include image segmentation, feature extraction, and hand gesture training and recognition. The main work of this paper includes:1. We analyse the advantages and disadvantages of skin color model based on YCbCr color space and motion information model, design and realize the combination of skin color and motion information of adjacent frames the adaptive update Gaussian modeling algorithm, get the better hand gesture segmentation.2. We propose a feature extraction and recognition algorithm based on convolutional neural network (CNN) and multilayer perceptron (MLP), and the method is compared with the traditional feature extraction method: compare with based on gradient (HOG) and based on local texture (LBP) feature extraction algorithm. Experimental results show that the recognition rate of using CNN and MLP with feature extraction algorithm is 97.86%, which is improved by about 5% compared with the traditional CNN, the training model is about 8 times faster than the other two algorithms, and the recognition rate than HOG+MLP higher than 8%; about 10% higher than the LBP+MLP.3. Use the proposed feature extraction algorithm, design and achieve the static gesture recognition algorithm of based on single camera and demo system, compared with other gesture recognition methods, such as based on histogram of oriented gradient (HOG) and support vector machine (SVM) feature extraction and based on local texture feature extraction (LBP) and support vector machine (SVM) method. The experimental results show that the recognition rate of the proposed algorithm is about 10% higher than that of HOG+SVM; it is about 12% higher than that of LBP+SVM.