面向人机交互的手势识别

发布时间：2018-04-12 21:14

本文选题：人机交互 + 手势识别　；参考：《华中科技大学》2016年硕士论文

【摘要】：随着信息技术的迅猛发展,计算机、智能电视等智能机器对于人们的生活愈发重要。将手势引入人机交互系统,使得人们可以和智能设备通过定义好的手势进行交流,这种改变将对日益趋向电子化的生活带来极大的便利。本文先介绍了一种基于手势识别的人机交互系统,系统由视频图像采集、视频图像分析、智能设备响应等三个部分构成。人脸检测、运动检测和手势识别共同组成了系统核心模块——视频图像分析。本文主要对面向人机交互的手势识别算法进行了研究并提出两种识别算法。基于轮廓特征的识别方法根据人脸检测算法检测到的人脸进行肤色特征提取;再利用肤色信息和运动信息完成肤色点检测并提取手势轮廓;通过对轮廓进行简化和拟合,得到近似手势的多边形以及角度,基于这些信息提取轮廓特征,从而实现多尺度多角度的手势识别。基于轮廓特征的识别方法在背景与肤色比较相近的情况下,误识率比较高,基于多分类器的手势识别算法克服了这一缺陷。算法主通过基于分割的跟踪算法进行人手跟踪,获取人手的大致轮廓以及位置,通过这些信息提取直立的人手图像块,最后提取该图像块的HOG特征,并通过多个分类器进行手势的识别。算法在复杂背景肤色相似背景均具有较高识别率。
[Abstract]:With the rapid development of information technology, computers, intelligent TV and other intelligent machines are becoming more and more important to people's lives.With the introduction of hand gestures into human-computer interaction systems, people can communicate with intelligent devices through well-defined gestures, which will bring great convenience to the increasingly electronic life.This paper first introduces a human-computer interaction system based on gesture recognition, which consists of three parts: video image acquisition, video image analysis and intelligent device response.Face detection, motion detection and gesture recognition constitute the core module of the system-video image analysis.In this paper, the gesture recognition algorithm for human-computer interaction is studied and two recognition algorithms are proposed.The recognition method based on contour feature is used to extract the skin color feature of the face detected by the face detection algorithm; then the skin color information and motion information are used to detect the skin color points and extract the contour of the gesture; the contour is simplified and fitted.The polygon and angle of the approximate gesture are obtained, and the contour features are extracted based on these information, and the multi-scale and multi-angle gesture recognition is realized.The recognition method based on contour features has a high error rate when the background is similar to the skin color. The multi-classifier based gesture recognition algorithm overcomes this defect.The algorithm uses segment-based tracking algorithm to get the rough contour and position of the human hand, extract the vertical human image block through these information, and finally extract the HOG features of the image block.Gestures are recognized by multiple classifiers.The algorithm has high recognition rate for complex background with similar skin color background.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】