基于肤色分割和统计模板匹配的手势识别人机交互系统

发布时间：2018-06-05 01:18

本文选题：单目视觉 + 肤色分割　；参考：《广东技术师范学院》2017年硕士论文

【摘要】：近年来,随着计算机领域技术发展迅猛,更加自然、高效的新型人机交互方式不断涌现。手势是人类的基本沟通方式之一,其符合人类的日常交流习惯。基于单目视觉技术,通过手势识别实现更符合人类交流习惯的人机交互,已成为人机交互领域的研究热点。目前,虽有不少手势肤色分割和手势识别算法被提出,但现有的算法在识别率、执行效率、以及实用性等方面仍然存在不足,有待改进。比如,大多数静态手势识别算法的复杂度高,而且在复杂背景或光照条件差的环境下难以获得理想的手势分割效果,进而导致手势识别率低下。针对这些问题,本文着重围绕手势肤色分割和静态手势识别这两个方面开展理论及应用研究,主要完成的工作及贡献如下:1.在分析相关技术的基础上,提出了一种综合多要素的手势肤色分割方法。该方法首先采用椭圆肤色模型对肤色进行初步分割,然后利用运动物体检测方法建立背景模型来排除背景中近似肤色的区域,进而结合人脸识别技术排除人脸肤色区域,最终分割出手势肤色区域。实验结果表明,本文提出的手势肤色分割方法在复杂背景或光照条件差的环境下能获得较好的手势分割效果。2.提出了一种简单有效的统计模板匹配算法,用以实现静态手势识别。首先,基于正态分布概率模型,利用采集得到的手势图像样本生成各种手势对应的统计模板特征;其次,利用模板特征定义手势图像之间的相似度,进而设计匹配判断规则对手势图像进行区分,以判断待识别手势图像对应的手势类别。针对11种手势的识别实验结果表明,本文提出的算法能获得高于93.5%的平均识别率,优于现有的同类算法。3.将前述提出的手势肤色分割和手势识别算法应用于人机交互,以C++为编程语言,结合MFC开发框架及OpenCV开源库,设计并实现了一个手势识别交互系统。该系统提供了11种手势,利用这些手势可以模拟鼠标和键盘操作,达到控制PPT、播放器等软件操作的目的。该系统界面友好,执行效率高,具有较高的通用性。
[Abstract]:In recent years, with the rapid development of computer technology, more natural, efficient and new human-computer interaction methods are emerging. Gesture is one of the basic ways of human communication, which accords with human daily communication habits. Based on monocular vision technology, it has become a research hotspot in the field of human-computer interaction to realize human-computer interaction which is more in line with human communication habits through gesture recognition. At present, although a lot of gesture color segmentation and gesture recognition algorithms have been proposed, the existing algorithms in recognition rate, execution efficiency, and practicability are still insufficient, and need to be improved. For example, most static gesture recognition algorithms have high complexity, and it is difficult to obtain ideal gesture segmentation results in complex background or poor lighting conditions, which leads to low gesture recognition rate. In order to solve these problems, this paper focuses on the theoretical and applied research of gesture skin color segmentation and static gesture recognition. The main work and contributions are as follows: 1. Based on the analysis of related techniques, a new method of gesture skin color segmentation is proposed. In this method, the skin color is initially segmented by using elliptical skin color model, and then the background model is established by moving object detection method to exclude the region of approximate skin color in the background, and then the skin color region of face is excluded by combining face recognition technology. Finally, the skin area of the gesture is segmented. The experimental results show that the proposed skin color segmentation method can achieve a better result of gesture segmentation under complex background or poor illumination conditions. A simple and effective statistical template matching algorithm is proposed to realize static gesture recognition. Firstly, based on the probability model of normal distribution, the statistical template features of various gesture images are generated by using the collected gesture image samples. Secondly, the similarity between gesture images is defined by template features. Then the matching judgment rules are designed to distinguish the gesture images to judge the corresponding gesture categories of the gesture images to be recognized. The experimental results of 11 hand gestures recognition show that the proposed algorithm can achieve an average recognition rate of more than 93.5%, which is better than the existing similar algorithm .3. The algorithm of gesture skin color segmentation and gesture recognition is applied to human-computer interaction. An interactive system of gesture recognition is designed and implemented by using C as programming language combined with MFC development framework and OpenCV open source library. The system provides 11 gestures which can be used to simulate mouse and keyboard operations to control PPTs, players and other software operations. The system has friendly interface, high execution efficiency and high generality.
【学位授予单位】：广东技术师范学院
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】