摘要：本文以带电作业机器人的遥操作为研究背景，设计了基于 Kinect 手势识别的人机交 互系统。首先利用 Kinect 体感设备采集并分割操作者的手部图像，采用几何不变矩 Hu 矩作为描述手部轮廓的特征，使用支持向量机和正态贝叶斯的机器学习方法分别对多种 静态手势样本进行训练，并对两种学习器的训练结果进行比较，实现了以手掌为模型的 静态手势识别；然后利用 Kinect Studio 和 Visual Gesture Builder 工具对操作者的手部动作 进行标记和训练，并对训练模型进行了分析验证，实现了以手部运动姿势为模型的动态 手势识别；最后设计了手势交互的人机界面，并对手势语义进行定义，实现了一个基于 Kinect 手势识别的人机交互系统。 

关键词  手势识别  人机交互界面  机器学习 Kinect

Title Human-Machine  Interaction  System  of  Tele-Manipulator  based on Kinect Gesture Recognition                      

Abstract：This study focus on the research background of the teleoperations of the live-linemaintenance robot, the human-machine interaction (HMI) system based on Kinect gesture recognition is designed. Firstly, the hand image of the operator was collected and segmented by Kinect somatosensory device. Then, the geometric invariant moments are used to describe the characteristics of the hand contours. Using the support vector machine and the normal Bayesian machine learning method, the training results of the two algorithms were compared to realize the static gesture recognition with the palm of the hand. Furthermore, the Kinect Studio and the Visual Gesture Builder tool were used to mark and train the hand movements of the operator. The training model is analyzed and verified, and the dynamic gesture recognition based on the hand movement posture is realized. Finally, the man-machine interface of gesture interaction is designed and the gesture semantics is defined to realize a man-machine interaction system based on Kinect gesture recognition.

Keywords  gesture recognition    human-machine interface    machine learning   Kinect

目 次

1 绪论 1

1.1 研究背景与意义 1

1.2 国内外研究现状 1

1.3 本文研究内容及安排 2

2 手势识别人机交互系统概述 4

2.1 人机交互系统设计概述 4

2.2 手势识别的流程简介 4

2.3 Kinect体感设备简介 5

2.4 本章小结 7

3 基于机器学习算法的静态手势识别 8

3.1 静态手势分割 8

3.2 静态手势特征提取 9

3.3 静态手势识别