CASIA OpenIR  > 毕业生  > 硕士学位论文
实时人手检测、跟踪与手势识别技术研究
其他题名Real-time Hand Detection, Tracking and Hand Gesture Recognition
周代国
学位类型工学硕士
导师王阳生
2009-06-02
学位授予单位中国科学院研究生院
学位授予地点中国科学院自动化研究所
学位专业计算机应用技术
关键词人机手势交互 Adaboost 方向梯度直方图 条件概率密度传播 隐马尔科夫模型 Gesture Based Human Computer Interaction Adaboost Histogram Of Orientation Gradient Condensation Hidden Markov Model
摘要伴随着软硬件技术的飞速发展,电子计算机已逐渐深入到人们生产生活的方方面面,这使得人机交互成为一个在理论和实践上都具有重要意义的研究课题。当前的人机交互主要依赖于键盘、鼠标以及一些图形化界面,这些传统方式难以满足日益多样化的应用需求,具有自然、直观、友好等特点的新型交互方式成为一种必然趋势。基于视觉的手势交互技术具有以上特点,目前在数字娱乐、虚拟现实、手语理解、远程控制等领域中有着广泛而重要的应用。 鉴于实时手势交互技术所具有的重要学术价值和广泛应用前景,本文选择自然背景下实时的人手检测、跟踪以及动态手势识别作为论题进行了深入的研究,最终实现了一个完整的手势识别系统。本文的主要内容概述如下: 1. 将人脸检测中广泛采用的Adaboost算法用于人手的检测,并用方向梯度直方图(HOG)特征替换传统的Haar特征进行训练和检测,实现了高效的人手检测。 2. 根据人手检测结果自动初始化跟踪算法,并利用最佳初始轮廓构造出一个线性肤色分类器。该分类器针对特定人建立,计算量少并具有较好的效果,还可以在后期跟踪过程中根据肤色提取效果进行参数调整,以适应一定的环境变化。 3. 利用B样条曲线建模人手,采用Condensation算法跟踪轮廓的刚体运动(平移、旋转及缩放),再采用启发式扫描方法确定各个手指轮廓的角度和长度,对运动自由度很高的人手轮廓进行了实时高效的跟踪。 4. 利用隐马尔科夫模型(HMM)实现了动态手势识别,并在最后将检测、跟踪和识别融合成一个完整的系统。
其他摘要Rapid development of software and hardware enables wide application of electronic computer in various aspects of life, which makes the study on Human-Computer Interaction practically and theoretically significant. Traditional tools of HCI, such as keyboard, mouse and graphic interface, cannot meet the demand of more and more diverse applications. Therefore, it is an inevitable trend to develop new model of interaction featured by simplicity, immediacy and user-friendliness. In this regard, vision-based hand gesture interaction technology satisfies current needs and is now extensively used in Digital Entertainment, Virtual Reality, Sign Language Recognition, Remote Control and other areas. In view of academic value of hand gesture based interaction technology and its prospect of wide application, this thesis aims to make an in-depth study on the hand detection, hand track and dynamic hand gesture recognition of real-time against natural background, and to work out a dynamic gesture recognition system. The thesis includes following main contents. 1. Adaboost, a widely used algorithm in the detection of human face, is employed to detect human hands. And the feature called Histogram of Orientation Gradient (HOG) is used to replace traditional Haar feature. Combination of HOG and efficient Adaboost ensures satisfactory effect of human hand detection. 2. Use the result of hand detection to initialize the hand tracker automatically. Construct a linear skin classifier from the optimal initialized hand contour. The skin classifier is highly individual-specific, it claims relatively fewer computational efforts but works out good effect. Moreover, it allows parameter adjustment based on skin extracting effect in later tracking stage, which makes it adaptive to changes of environment. 3. Real-time tracking of human hand contour with high degree of freedom is achieved by combining Condensation and a heuristic scanning method. Condensation is used to determine the rigid motion of the whole hand, and heuristic scanning method is used to find the angle and length of each finger. 4. Trajectory of human hand is recognized with Hidden Markov Chain (HMM). And a system combining hand detection, contour tracking and gesture recognition is built.
馆藏号XWLW1419
其他标识符200628014629090
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/7489
专题毕业生_硕士学位论文
推荐引用方式
GB/T 7714
周代国. 实时人手检测、跟踪与手势识别技术研究[D]. 中国科学院自动化研究所. 中国科学院研究生院,2009.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
CASIA_20062801462909(1295KB) 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[周代国]的文章
百度学术
百度学术中相似的文章
[周代国]的文章
必应学术
必应学术中相似的文章
[周代国]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。