| 英文摘要 | As a rising industry, compute game is gaining more and more attention fromall over the world. Interactive technique is an important part of computer gamefor control. Along with the development of compute techniques, computer game is becoming so complicated and multiplex that traditional human computer interactive technique can't meet requirement of this situation. At the same time,consumer also needs to interact with computer game more naturally and intelligently, including voice, vision, intelligent sensors, and so on.Among all methods regarding natural human computer interactive research, computer vision based interactive method draws a lot of attention because it has the advantage of nature and convenience. In this article, our key research is how to use the vision information to realize the interaction based on gesture.By investing and combining current game control ways, we propose novel methods for two classic applications. The main contributions of this article are as follows: First, we propose a new algorithm for estimating head pose under small rotation. It includes two steps: first, 2D model based key point location algorithm is used to obtain the key points of human face in current image, and then, 3D head model is used to estimate head pose. There are many 2D model based key point location algorithms proposed by other researchers currently. Some of those algorithms are too slower, and some of them are sense to illumination. In this article, we propose an improved active appearance models. Keeping the high speed, our algorithm is more robust to illumination and more precision than original algorithm. Second, we propose an innovative algorithm for estimating head pose under large rotation. It integrates detection, tracking and recognition to a simple framework. By introducing Markov model and utilizing multi-info from previous steps, this algorithm decreases its complexity of computing. Its sensitivity to illumination also reduced by employing anti-light step. Third, we propose a hand gesture recognition method for driver simulation. Tracking fists robustly in complex background is a challenging issue for researchers in this field. We propose a muti-information fusion based tracking method under Bayesian network to sort out this problem. Then geometric relationship between two fists is exploited to estimate driver gesture. This algorithm keeps the character of high running speed, and tracks location of fists accurately as well. Forth, according to the research of two interactive techniques on gesture in this article, two kinds of pioneering application for computer game are proposed. I)Based on head pose estimation algorithm, we design a view control plug-in for 3D game. II)Based on hand gesture interactive technique, we design a racing carcontrol system. |
修改评论