The Research On The Key Technology Of Vison-Based Hand Gesture Recognition System

Artificial intelligence, as a new wave of technology, is changing the world. And the human-machine interaction as artificial intelligence interface is getting more and more attentions of researchers all over the world. Of all the existing human-machine interaction technologies, the hand gesture recognition is one of the most popular one, and with its closest to human communication way, will certainly become the indispensable part of human-machine interaction in the future. Hand gesture recognition system can be divided into three modules including hand gesture modeling, hand gesture analysis and hand gesture recognition, and has the function of static hand gesture recognition and dynamic gesture recognition.This paper mainly researches the key technology of hand gesture recognition based on computer vision, and respectively establishes the hand gesture recognition system for static and dynamic hand recognition. In the process of recognizing the static hand gestures, establishes a hand gesture library with 3000 hand gesture examples of 10 different hand gestures, and designs a set of preprocessing module for the recognition system, then extracts Hu moments and other geometric features to train the static gesture classifier based on support vector machine, finally finds the best parameters of kernel function through the grid traversal to get above 90% recognition rate in the static hand examples,and 80% recognition rate in the real-time testing environment, at last compares the effects of different features and parameters to the static gesture classifier.In the dynamic hand recognition system,7 different dynamic hand trajectories are collected. By narrowing the scope of choosing the optimal path and designing the new similarity measure between patterns,24.8% calculating time is saved. The dynamic hand gesture recognition system includes the static gesture recognition technologies. The system needs to recognize the static hand target in the input signal firstly, and then tracks and obtains the dynamic gesture trajectory in the form of time series based on Kalman filter model. It transforms the original trajectory consisting of a series of points to the counter clockwise angles between the horizontal axis and the straight line that joins the adjacent points of trajectory, then uses the improved dynamic time warping algorithm to identify the dynamic hand gestures.In conclusion, this paper introduces the whole process of hand gesture recognition and the platform of system software and hardware. In a real-time environment, the algorithms used in the static hand gestures system and dynamic hand gesture recognition are tested and results show that the hand gesture recognition technologies based on computer vision in this paper can basically meet the need of gesture recognition, and has great application prospect and research value, which will become one of the main methods of the human-machine interaction interface in the future.