中文摘要：

随着更多照相和视频捕捉设备的涌现,对于非接触式手势命令的识别提出了很高的需求.本文针对这一趋势,依据实验环境和实际应用的需要,提出了一种在基于OpenCV2.2视觉库和Visual Studio C＋＋来实现的Haar的矩形特征提取并充分利用Adaboost的学习分类模块来实现对特定握拳手势的实时识别和精确定位.本方法使用的迭代算法将弱分类器训练组合为强分类器,经过基于正负样本图片的过程后,所得的级联分类器首先可以根据摄像头捕捉的视频中的实时手势位置,能够识别手势的类型并进行对应画笔轨迹的跟踪实验,并且通过具体的算法,在已识别的各个位置中,实现了去除可能的误差位置,从而使得画笔的轨迹更为流畅.根据统计,所进行的实验对室内环境下握拳手势的识别率可以达到90％,取得了良好的效果.

英文摘要：

The more camera and video capture devises are put in use,the higher request for non-touch gesture recognition becomes.In this article,based on the OpenCV2.2 visual library and Visual Studio C ＋ ＋,the authors implement the module of extracting Haar features and multi-scales classification with Adaboost to exploit special gesture recognition to meet this trend.Using the recognition and location of clench fist gesture in real time,the authors have completed the tracking of brush track through the location of gestures,and also worked out an algorithm to decrease the errors in detected points to make the brush go smoothly.According to the experiments,under the indoor condition,the recognition rate can achieve as good as 90％.