CASIA OpenIR  > 毕业生  > 硕士学位论文
任意手势的跟踪与识别技术研究
Alternative TitleVision-based Arbitrary Hand-shape Tracking and Recognition
石磊
Subtype工学硕士
Thesis Advisor王阳生
2011-05-30
Degree Grantor中国科学院研究生院
Place of Conferral中国科学院自动化研究所
Degree Discipline计算机应用技术
Keyword人机手势交互 任意手势跟踪 手势识别 Svm 隐马尔科夫模型 Gesture Based Human Computer Interaction Arbitrary Hand-shape Tracking HAnd Postures And Gestures Recognition Support Vector Machine Hidden Markov Model
Abstract随着电子计算机的普及,越来越多的人使用计算机来进行工作和娱乐,传统的鼠标键盘作为主流人机交互媒介已经有数十年之久。然而,这种交互方式难以满足日益多样化的应用需求,一种自然、和谐的交互方式必将成为一个新的趋势。基于计算机视觉的手势交互是一种新颖的交互方式,鉴于其广泛的应用前景,众多专家致力于这方面研究,已经取得了很多成果。 本文选择任意手势下的跟踪与识别技术为题进行了深入的研究,旨在实现一种自然、直接的人机交互接口,不对用户进行过多限制。本文在手势检测、跟踪与识别等方面均进行了深入探讨,最终提出了一个手势交互原型系统。本文的主要内容概括如下: 1.采用梯度直方图作为特征,AdaBoost算法作为分类方法,实现了实时人手检测。 2.对传统的MeanShift跟踪算法进行改进,采用在线学习算法MIRA(Margin Infused Relaxed Algorithm)对目标进行实时更新,训练一个SVM模型,作为跟踪目标,实现了对任意变化手势的跟踪。 3.融合了光流法与Camshift算法,采用肤色特征角点作为跟踪目标,利用稀疏点光流计算进行跟踪,并结合上下文进行运动预测,结合Camshift整体跟踪结果对跟踪点进行实时更新,实现了对任意手势的跟踪,排除了其他大面积肤色物体的干扰。 4.利用归一化的傅里叶描述子表述静态手势特征,采用SVM分类方法进行识别,并且使用隐马尔科夫模型对动态手势进行识别。最终提出一个完整的手势交互原型系统,并以手势PPT播放系统为例对系统进行了验证。
Other AbstractAlong with the popularization of computers, more and more people choose a computer as an aid in their work and entertainment. As a mainstream human computer interaction medium, the keyboard and mouse has been used for several decades. However, this interaction pattern cannot meet increasingly diversified application requirement. So a more natural and harmonious pattern must be a new trend. Computer vision based hand gestures HCI is being a hot area, as it has a wide application prospect, on which many researcher study and gain some achievement. Aim to realize a natural and direct interface and not to impose much restrictions on uses, this thesis choose vision-based arbitrary hand-shape tracking and recognition to conduct in-depth study, including hand detection, tracing and recognition. At last, the thesis presents a hand gestures interaction prototype system. The thesis includes the following main contents: 1.The Histogram of Orientation Gradient(HOG) is used as the feature, and AdaBoost algorithm is utilized as the classification method. The combined method realizes a real time hand detector. 2.The traditional MeanShift tracking algorithm is improved, using an online learning algorithm named Margin Infused Relaxed Algorithm, training a SVM model as the real time updating target. It realizes arbitrary hand-shape tracking. 3.Blending the optical flow and Camshift as a whole, using several skin feature corner points as the tracking target, utilizing the sparse points optical flow method, the thesis realizes tracking of hand. A motion prediction mechanism is used, combining Camshift to update the tracking target points, to get rid of the interference of other big skin-like objects. 4.Hand postures are recognized using normalized Fourier descriptors and Support Vector Machine, and hand gestures are recognized using Hidden Markov Models. Finally, an hand gestures HCI prototype system is presented and is verified using hand gesture PPT controller as the example.
shelfnumXWLW1652
Other Identifier200828014629083
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/7555
Collection毕业生_硕士学位论文
Recommended Citation
GB/T 7714
石磊. 任意手势的跟踪与识别技术研究[D]. 中国科学院自动化研究所. 中国科学院研究生院,2011.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_20082801462908(2885KB) 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[石磊]'s Articles
Baidu academic
Similar articles in Baidu academic
[石磊]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[石磊]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.