基于视觉的手势识别及其交互应用研究

发布时间：2018-06-11 12:42

本文选题：目标检测 + 手势分割　；参考：《南京理工大学》2017年硕士论文

【摘要】：科技进步使人机交互方式朝着更加自然、人性化的方向发展,传统的交互方式已不能满足人们的需求。近年来增强现实和虚拟现实技术发展迅速,推动了基于手势识别的交互技术的发展,除此之外,手势识别在无人机控制、智能家居和手语识别等众多领域都有广泛应用,在此背景下,本文对手势识别算法进行研究,并最终模拟鼠标功能,实现了单目视觉下自然人手的人机交互。本文所实现交互系统由手势分割、手势跟踪、手势识别、系统实现等模块组成。在手势分割模块中,针对固定阈值的肤色分割方法不能适应实际复杂多变环境的问题,提出了对人手进行现场肤色建模,并利用此模型进行后续手势的分割,实验结果显示能有效地从复杂背.景中分割出手势。在手势跟踪模块,使用核相关滤波器跟踪手势目标,针目标跟踪丢失的问题,提出了2种目标再检测机制。在跟踪前需要对目标初始化,本文使用支持向量机和滑动窗口检测人手,但滑动窗口遍历整幅图片,带来巨大的时间开销,针对人手的运动特性以及背景静止的特点,提出了检测之前先利用改进的帧间差分法检测运动区域,缩小检测范围,该方法使检测区域减少到原来的四分之一,显著提高了检测速率。在手势识别模块,使用傅里叶描述子作为手势特征,选择k最近邻法对静态手势进行识别,在动态手势识别中,提出了更简洁的统计计数识别方法,此算法完全满足系统实时性要求。系统实现模块调用应用程序接口函数实现对鼠标的模拟,并利用MFC开发了一个对话框程序,对手势识别结果进行直观的展示。
[Abstract]:The progress of science and technology makes the human-computer interaction more natural and humanized. The traditional way of interaction can not meet the needs of people. In recent years, the rapid development of augmented reality and virtual reality technology has promoted the development of interactive technology based on gesture recognition. In addition, gesture recognition has been widely used in many fields, such as UAV control, smart home and sign language recognition, etc. Under this background, this paper studies the gesture recognition algorithm, and finally simulates the mouse function, realizes the man-machine interaction of natural hand under monocular vision. The interactive system is composed of gesture segmentation, gesture tracking, gesture recognition, system implementation and so on. In the hand gesture segmentation module, aiming at the problem that the fixed threshold skin color segmentation method can not adapt to the actual complex and changeable environment, a skin color modeling method is proposed for the human hand, and the following hand gesture segmentation is carried out using this model. The experimental results show that it is effective from the complex back. Cut out the gestures in the scene. In the hand gesture tracking module, the kernel correlation filter is used to track the gesture target, and the problem of missing needle target tracking is discussed. Two kinds of target re-detection mechanisms are proposed. It is necessary to initialize the target before tracking. Support vector machine and sliding window are used to detect the human hand, but the sliding window traverses the whole picture, which brings huge time cost, aiming at the movement characteristics of the hand and the static background. Before detection, the improved inter-frame differential method is used to detect the motion area and reduce the detection range. This method reduces the detection area to 1/4, and improves the detection rate significantly. In the gesture recognition module, the Fourier descriptor is used as the gesture feature, and the k-nearest neighbor method is selected to recognize the static gesture. In dynamic gesture recognition, a more concise statistical counting recognition method is proposed. This algorithm fully meets the real-time requirement of the system. The system realizes the simulation of mouse by calling the application program interface function, and develops a dialog program with MFC to show the result of gesture recognition intuitively.
【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】