基于矩形特征和改进Adaboost的手势检测  

Hand Detection Based on Rectangle Features and Improved Adaboost

作　　者：张秋余[1] 姚开博[1] 吴佩莉[1] 

机构地区：[1]兰州理工大学计算机与通信学院,兰州730050

出　　处：《计算机工程》2008年第14期176-178,共3页Computer Engineering

基　　金：甘肃省自然科学基金资助项目(ZS021-A25-018-G)

摘　　要：为了实时、精确地从视频流中检测和识别出特定手势,提出一种矩形特征描述手势,给出快速计算方法和手势类可分离性的评价方法。为了避免分类器的过度训练问题,提出一种基于此方法的改进的Adaboost算法。实验结果表明,矩形特征能够产生可靠的检测器,对手势的姿态变化较敏感。在摄像头实时捕获视频中,其检测手势实时性较好,对复杂背景和噪声有较强的适应能力,当手势旋转角度小时,正确检测率可以达到95%以上。In order to detect and recognize hand gesture from video sequence in real time accurately, a set of rectangle features are used to describe the hand characteristics, and the fast calculation methods of their features and the evaluation ways of the separability of hand gesture class are given. The Adaboost algorithm is improved to deal with the excessive training. Experimental results show that rectangle features can obtain the reliable detector and has a good real-time performance and adaptive capacity in complex backgrounds, but it is more sensitive to the changes of hand gesture. When hand gesture rotates small angle, the detection rates are 95% above.

关 键 词：矩形特征 改进Adaboost 复杂背景 手势检测 手势识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]