Studies On Hand Gesture Tracking And Recognition Algorithm Based On Computer Vision

The hand gesture recognition based on computer vision is a new method for Human-Computer Interface, and it has important theoretical value and application prospect. This paper studies the algorithms of hand gesture tracking and recognition and realizes the system under Microsoft Visual C++ 6.0. Besides, based on the algorithm which is about recognizing ten common hand gestures inputted from camera in real time, a simple gesture interaction system has been established to test the feasibility of the algorithm. The results of experiment show that this system has strong rubustness in real-time tracking of the target gesture and high efficiency in recognition.By collecting gesture images from input video and pre-processing them, a sample database is established. As the testing samples are hand gestures and Hu Moments can overcome the uncertainty of rotation and scale better because of Geometric Moments feature which is not changed in pace with the image's changment in rotation, translation and scale, the Geometric Moments is used as gesture recognition feature in feature extraction.In multi-classification algorithms based on Support Vector Machine, the performance of the algorithms such as One-Against-Rest, One-Against-One and Directed Acyclic Graph used in gesture recognition has been analyzed and verified. Further research and simulation on parameter optimization of SVM in gesture recognition has been accomplished. The results of experiment show that One-Against-Rest RBF kernel function is ideal in gesture recognition. Besides, to the condition of wrongly classified that one sample belongs to several classes, multi-classification algorithms based on SVM and Posterior Probability is used in gesture recognition and simulationally tested. Taking the probability as the output of gesture classify can lessen the misclassification and finally the recognition rate can reach 98.9%.In the algorithm application, the emphasis is mainly on simulation experiment on self-defined ten common numerical hand gestures. To operate word software by gestures input from camera, a number-input program has been established based on VC++6.0 to make use of gesture recognition in human-computer interaction.The innovations of this paper are as follows, firstly, an algorithm based on the combination of Hu Moments and One-Against-Rest RBF kernel function is applied in gesture recognition and simulated, and the results are fairly good. Secondly, to the condition of wrongly classified that one sample belongs to several classes, multi-classification algorithms Posterior Probability is used in gesture recognition to raise recognition rate.