【作者】 江立；

【导师】 阮秋琦；

【作者基本信息】 北京交通大学 ， 信号与信息处理， 2006， 硕士

【摘要】 最近几年,随着计算机技术的迅猛发展,人与计算机的交互活动日益频繁,人机交互也成为人们日常生活的一个重要组成部分。依靠传统的交互方式,用户通过键盘、鼠标向计算机输入信息,这种方式的弊端在于用户不能以习惯的方式(如手势、语音)与计算机进行交互;而现代的交互方式冲破了人机通信的瓶颈,充分体现了以人为本的思想,通过手势、语音等方式实现人机交互,是一种多媒体、多种模态的交互技术。手势是一种自然、直观、易于学习的人机交互手段,与鼠标相比,手势不但提供了更加丰富的空间信息,而且自然舒适符合用户的交互习惯。手势的识别就是根据用户的手势识别手势的含义。本文描述一个基于数据手套的手势识别系统。本文采用中科院研发的CAS-GLOVE型数据手套,针对传感器的特性,将传感器的原始数据转换为角度数值,提高了系统的精确度和识别率。本文分析了手形的几何关系,建立了虚拟手的模型,由数据手套的数据接口获取各指节的曲伸角度,建立手势标准样本库,实现了基于BP神经网络的手势识别方法,用手势标准样本加以训练,使其具备识别手势的功能。本文还提出了基于决策树的手势识别方法,实现了实时识别功能。该算法非常简单,识别速度快,识别率高,其缺点是容易造成误差的累积,使决策树中离根节点较远的样本识别率较低,而且无法进行拒识别。更多还原

【Abstract】 Recently, with the rapid development of the technology of computer, the interaction between human and computer is more and more continual, and it becomes an important part of our daily life. By the traditional way of interaction, the consumers input the information with the board or the mouse. The abuse of it is that the consumers cannot communicate with the computer with their habitual modes, such as gesture, voice etc. But the modern modes of the interaction break the choke point of the interaction between human and computer adequately represent the idea of“human beings are principal”. It is an interaction technology of multimedia and multimode to actualize the interaction between human and computer by gesture or voice.Gesture is a natural, intuitionistic and easy mode for the interaction between human and computer. Compared with the mouse, gesture not only supplies more plentiful space information, but also accord with our habit of the interaction which is much more spontaneous and convenient. Gesture recognition means that we can recognize the meaning according the consumer’s gesture. The paper describes a gesture recognition system based on data glove.We use the CAS-Glove developed by CAS. We transform the original data into the angle value according the characteristic of the sensors, and then the precision of the net-training can be effectively improved. This paper analyses the geometric relation of hand shapes. The model of virtual hand is constructed. The tortile angle data of each joint is got from the serial data port of the glove. The standard sample copy library is built. We achieve the gesture recognition with BP Neuron Networks. The network is trained by the standard samples, and it has the function of gesture recognition.We also put forward the method of gesture recognition based on Decision Tree, and achieve real time recognition. This arithmetic is simpler, costs less time, and has a higher recognition rate. The disadvantage is that it causes the accumulation of error easily, so the recognition rate of the sample which is far from the root node is lower, moreover, it can’t reject the recognition.更多还原