【作者】 张鹏；

【导师】 刘云；

【作者基本信息】 青岛科技大学 ， 计算机软件与理论， 2010， 硕士

【摘要】 随着人机交互技术向着“以人为中心”的多媒体、多模式交互的方向发展,传统的基于键盘、鼠标的交互方式越来越显示出自身的局限性;将基于计算机视觉的手势识别方法融入到新一代人机交互模式中,成为完善人机交互手段的一种新的思路。本课题面向人机交互应用,针对复杂背景下交互手势的检测方法、手势的特征描述方法、手势识别方法做了深入研究,并在上述研究基础上选取虚拟现实环境为应用背景,建立了一个基于用户手势输入的虚实交互系统,实现了人机之间的友好交互。在手势检测方面,针对实际应用中背景复杂多变的特点,提出基于肤色分割前端优化的Viola-Jones手势检测方法。为排除光照强度变化的影响,在肤色分割模块中实现了在非线性转换YCbCr颜色空间中的手势建模。肤色分割模块的引入,有效地解除了复杂背景对手势检测问题的制约,在一定程度上降低了算法的误检率。在算法实现环节分别通过复杂背景下测试、分类器性能测试、实时性能测试对上述方法进行验证,实验结果表明本文方法对复杂背景下的手势检测较为鲁棒,表现出良好的整体性能。在手势识别方面,选取Hu不变矩作为手势的特征描述,提出结合Hu矩特征和支持向量机(SVM)分类的手势识别算法。Hu矩特征不易受噪声干扰,对检测手势在尺度、旋转角度等方面的变化有较强的适应性。实验环节中上述识别方法在测试样本集上取得了理想的识别率,从而进一步证明选取SVM用于特征分类,可有效地解决手势识别研究中面临的小样本、分类模型推广能力差及参数难以优化等问题。最后在上述研究基础上,将手势检测算法与识别算法结合,实现了对输入手势的自动识别,并选取虚拟现实环境为应用背景,在Visual C++ 6.0环境下建立了一个以用户手势为输入的虚实交互系统,利用对用户手势的识别结果控制虚拟环境中的物体,从而实现了人与虚拟环境的实时、友好交互。更多还原

【Abstract】 While Human-Computer Interaction (HCI) technology has developed into the human-centered, multi-mode and multimedia-supported stage, the traditional interaction modes based on mouse and keyboards increasingly show their limitations. Consequently, the application of vision-based gesture recognition to HCI with the advanced interactive models will provide a new research idea, which could further improve the modality of HCI.For the purpose of application in HCI, this thesis further studied gesture detection method in the complex backgrounds, gesture feature extraction method and gesture recognition method. On the basis of these studies, virtual environment was regarded as the application background, a Virtual-Reality Interaction system based on gesture recognition has been developed and the friendly interaction between users and computer was finally realized.On the stage of gesture detection, cluttered backgrounds were taken into account; the Viola-Jones gesture detection method with skin-color segmentation optimization module was presented. To reduce the interference of illumination variation, gesture modeling of the skin-color module was carried out in the Nonlinear Transformed YCbCr color space. The introduction of skin-color segmentation module could overcome the restriction of complex backgrounds effectively in gesture detection process and the false alarm rate was correspondingly reduced. In the experiment course, detection in complex background testing, classifier performance testing and real-time performance testing were carried out. The experiment results showed the proposed gesture detection method has a strong adaptability in complex backgrounds and a good real-time performance.On the stage of gesture recognition, Hu invariant moment method was applied for the feature extraction of gesture, which is less sensitive to noise and has a stronger adaptability with the scale and rotation variation of detected gesture. Furthermore, a new gesture recognition method based on Hu Moments and Support Vector Machine classification algorithm was presented. In the experiment course, the proposed recognition method achieved the desired results on testing sample sets. The experiment results further proved that SVM classification algorithm could effectively solve the problem of scared samples and weak generalization of the classification model in gesture recognition research.On the part of application, the automatic gesture recognition was firstly realized on basis of the proposed gesture detection and recognition methods. Furthermore, a Virtual-Reality Interaction system was developed with VC++ 6.0. This system could take the gesture of users as input commands to control movements of the virtual plane in a virtual enviroment, which realized friendly interaction between users and computer.更多还原