手势识别图像匹配技术的应用与研究  

The Research and Application of Gesture Recognition Image Matching Technique

作　　者：王庆平[1] 郭思哲[2] 和丽芳[1] 李涛[1] 

机构地区：[1]昆明理工大学城市学院,昆明650051 [2]昆明理工大学知识产权发展研究院,昆明650093

出　　处：《价值工程》2016年第19期185-187,共3页Value Engineering

基　　金：国家自然基金(51204077)

摘　　要：手势识别的前提条件是场景中手势的正确定位,通过皮肤颜色这一特点可以实现手势的定位。运用和比较了基于YCbCr色彩模型和BP神经网络两种肤色检测的方法。然后,通过帧间差分动态检测的方法,检测出静态的有用手势,将运动的手势判定为无用手势。最后,通过模板匹配找出被识别图像匹配度最大的模板,并认定它的手势为被识别图像的手势。在模板匹配中,为了消除手势的抖动,减小抖动对识别的影响,引入了自适应,从而使匹配更加精确。Gesture recognition is a prerequisite for the correct positioning of the scene gestures. In this paper, we use the characteristics of skin color to achieve the positioning of gestures. Two color detection methods are used and compared based on YCbCr color model and BP neural network. Then, we judge static gestures as the useful gestures while the movement gestures as useless gestures by the method of the dynamic detection of frame difference. Finally, we find the template which gets the largest matching score to match the image to be identified as the best template, and judge its gesture as the image＇s gesture. In template matching, a kind of adaptive is introduced in order to eliminate hitter gestures and reduce the impact of jitter on the identification, thus make more accurate matching.

关 键 词：YCbCr色彩模型 BP神经网络 帧间差分 模板匹配 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]