【作者】 彭毅； 包恺俊； 张自强； 李传江；

【Author】 PENG Yi;BAO Kaijun;ZHANG Ziqiang;LI Chuanjiang;College of Information, Mechanical and Electrical Engineering of Shanghai Normal University;

【机构】 上海师范大学信息与机电工程学院；

【摘要】 随着人机交互技术的发展,手势识别这一包含了最多信息量的人体语言凭借着其自然、友好、有效等天然优势越来越受到人们的关注。本文真对传统基于视觉的手势识别技术易受光照条件和复杂背景影响的缺点,使用kinect传感器作为手部信息采集的工具。在对Kinect深度传感器采集到的数据进行一系列处理后,根据阈值法将手势区分为静态手势和动态手势,然后针对这两种不同的手势分别采用了不同的识别算法进行识别。对于静态手势以手部轮廓的Hu矩作为特征分别利用利用BP算法和GA改进后的BP算法进行学习识别,对于动态手势采用DTW算法对其运动描述子序列进行模板匹配。两种手势的识别均取得了较为理想的效果。更多还原

【Abstract】 With the development of humancomputer interaction technology, gesture recognition as a body language contains most of the information attract more and more attention for it is advantages, just like, natural, friendly, and efficient.This paper using the Kinect sensor to collect hands information, in view of traditional gesture recognition based on vision is easily affected by illumination condition and complex background disadvantage. A series of processing after the Kinect depth sensor collected data, then using threshold method to distinguish the static gestures and dynamic gestures, these two different gestures were identified by different algorithms. For static hand gestures, using Hu moment of hand contour as features to training and recognition respectively BP neural network and GA improved BP algorithm. For dynamic gestures, using Dynamic descriptor as feature to emplate matching by DTW algorithm. These two different types of gestures have achieved satisfactory recognition effect.更多还原