基于KNN算法的空间手势识别研究与应用
发布时间：2018-11-07 08:49
【摘要】：近些年互联网和计算机硬件发展迅速,为互联网其它领域的发展打下了坚实的基础,在理论研究方面机器学习和人工智能均是研究的热点,应用方面VR和AR等虚拟现实设备也进一步成熟起来,很多公司已经推出了面向市场的产品线,虽然在人工智能方面还不能做到容错率媲美人类,但是现如今我们的生活已经被不少人工智能的产品所包围比如:手机中的手写识别与微信中的语音识别等应用场景。这些智能场景的应用是依托于计算机分类学习算法的大力发展,基础的分类方法如KNN和SVM分类,经过多年的研究和应用已经取得了满意的分类精度和快速的分类速度,KNN在数据量少的场景下是一个经典的分类解决方案,同时SVM也广泛的应用在二分类问题领域互相之间有一定的互补性。如上所示,人工智能的发展给用户带来了便捷的操作的同时极大的提升了用户体验,并且在其他领域也有着非常大的发展潜力。VR和AR等虚拟现实设备以及应用也在过去的两年中获得广泛关注和快速的发展,硬件的升级已经可以支撑其让虚拟现实走进我们的生活,HTC的虚拟头盔VIVE以及微软的AR眼镜Hololens从发布以来就吸引了无数的目光。本文通过对国内外研究现状进行分析以及对当下虚拟现实领域中的各种应用实践,在深入研究机器学习分类算法和HTC的VIVE设备应用系统的基础之上,提出基于KNN算法的空间手势识别研究并且在VR环境下进行了实际应用。研究内容包括:(1)探究如何解决VR中的人机交互需求,如何缩小用户和设备之间沟通的距离,让用户更加自然的沉浸到虚拟现实设备所构建的场景中去,在HTC的VIVE设备中通过一个类似于鼠标的控制器来让用户控制其实现操作,但是这种交互操作时常会干扰到用户的正常使用,体验并不完善,探究在此硬件设备的基础上使用机器学习分类算法搭建手势识别场景。(2)深入研究KNN和SVM分类算法原理和运行机制,对其基本的应用思想和代码结构进行深入理解,在此基础上分析不同分类算法的优缺点,并根据所要构建的VR手势识别应用,分析其适用度,做出算法的选择,并从算法实现复杂度和算法性能两个方向来尝试优化。(3)对KNN和SVM算法进行编程实现,使用标准数据集测试其效率和性能,通过对相关参数的调优,如KNN算法中K值的选取和SVM算法中的相关参数的调试来实现算法优化。最终通过对KNN和SVM进行组合应用,取长补短,实现分类效果的优化,并得出优化分析报告。(4)掌握当下VR前沿技术,在HTC的VIVE平台中利用Unity游戏开发引擎构建手势识别应用场景以及使用Python语言实现相关服务器数据传输和回传结果的功能模块,最终实现的手势识别应用。
[Abstract]:In recent years, the rapid development of Internet and computer hardware has laid a solid foundation for the development of other areas of the Internet. Machine learning and artificial intelligence are hot spots in theoretical research. Virtual reality devices such as VR and AR have matured further in applications, and many companies have launched market-oriented product lines, although in artificial intelligence fault tolerance is not comparable to that of humans. But now our lives are surrounded by artificial intelligence products such as handwritten recognition in mobile phones and voice recognition in WeChat. The application of these intelligent scenes depends on the development of computer classification learning algorithms. The basic classification methods, such as KNN and SVM classification, have achieved satisfactory classification accuracy and fast classification speed after many years of research and application. KNN is a classical classification solution in the case of small amount of data, and SVM is also widely used in the field of two classification problems, which is complementary to each other to some extent. As shown above, the development of artificial intelligence not only brings convenience to users, but also greatly improves the user experience. Virtual reality devices and applications, such as VR and AR, have also received extensive attention and rapid development in the past two years. Hardware upgrades have enabled virtual reality to come into our lives, with HTC's virtual helmet VIVE and Microsoft's AR Glass Hololens attracting a lot of attention since its launch. Based on the analysis of domestic and international research status and various application practices in the field of virtual reality, this paper deeply studies the machine learning classification algorithm and the VIVE device application system of HTC. Spatial gesture recognition based on KNN algorithm is proposed and applied in VR environment. The research contents include: (1) explore how to solve the need of human-computer interaction in VR, how to narrow the communication distance between users and devices, so that users can immerse themselves more naturally in the scene constructed by virtual reality devices. In HTC's VIVE device, a mouse-like controller is used to allow the user to control its implementation, but this interaction often interferes with the user's normal use and the experience is not perfect. On the basis of this hardware device, this paper uses machine learning classification algorithm to build gesture recognition scene. (2) deeply study the principle and running mechanism of KNN and SVM classification algorithm, and deeply understand its basic application thought and code structure. On this basis, the advantages and disadvantages of different classification algorithms are analyzed, and according to the VR gesture recognition application to be constructed, the applicability of the algorithm is analyzed, and the algorithm selection is made. And try to optimize the algorithm from two aspects of the algorithm implementation complexity and algorithm performance. (3) programming the KNN and SVM algorithms, using standard data sets to test its efficiency and performance, by tuning the relevant parameters. For example, the selection of K value in the KNN algorithm and the debugging of the related parameters in the SVM algorithm are used to optimize the algorithm. Finally, through the combined application of KNN and SVM, we can learn from each other and optimize the classification effect. (4) mastering the current frontier technology of VR. In the VIVE platform of HTC, the application scene of gesture recognition is constructed by using Unity game development engine, and the function module of data transmission and return result of related server is realized by Python language. Finally, the application of gesture recognition is realized.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18;TP391.9
[Abstract]:In recent years, the rapid development of Internet and computer hardware has laid a solid foundation for the development of other areas of the Internet. Machine learning and artificial intelligence are hot spots in theoretical research. Virtual reality devices such as VR and AR have matured further in applications, and many companies have launched market-oriented product lines, although in artificial intelligence fault tolerance is not comparable to that of humans. But now our lives are surrounded by artificial intelligence products such as handwritten recognition in mobile phones and voice recognition in WeChat. The application of these intelligent scenes depends on the development of computer classification learning algorithms. The basic classification methods, such as KNN and SVM classification, have achieved satisfactory classification accuracy and fast classification speed after many years of research and application. KNN is a classical classification solution in the case of small amount of data, and SVM is also widely used in the field of two classification problems, which is complementary to each other to some extent. As shown above, the development of artificial intelligence not only brings convenience to users, but also greatly improves the user experience. Virtual reality devices and applications, such as VR and AR, have also received extensive attention and rapid development in the past two years. Hardware upgrades have enabled virtual reality to come into our lives, with HTC's virtual helmet VIVE and Microsoft's AR Glass Hololens attracting a lot of attention since its launch. Based on the analysis of domestic and international research status and various application practices in the field of virtual reality, this paper deeply studies the machine learning classification algorithm and the VIVE device application system of HTC. Spatial gesture recognition based on KNN algorithm is proposed and applied in VR environment. The research contents include: (1) explore how to solve the need of human-computer interaction in VR, how to narrow the communication distance between users and devices, so that users can immerse themselves more naturally in the scene constructed by virtual reality devices. In HTC's VIVE device, a mouse-like controller is used to allow the user to control its implementation, but this interaction often interferes with the user's normal use and the experience is not perfect. On the basis of this hardware device, this paper uses machine learning classification algorithm to build gesture recognition scene. (2) deeply study the principle and running mechanism of KNN and SVM classification algorithm, and deeply understand its basic application thought and code structure. On this basis, the advantages and disadvantages of different classification algorithms are analyzed, and according to the VR gesture recognition application to be constructed, the applicability of the algorithm is analyzed, and the algorithm selection is made. And try to optimize the algorithm from two aspects of the algorithm implementation complexity and algorithm performance. (3) programming the KNN and SVM algorithms, using standard data sets to test its efficiency and performance, by tuning the relevant parameters. For example, the selection of K value in the KNN algorithm and the debugging of the related parameters in the SVM algorithm are used to optimize the algorithm. Finally, through the combined application of KNN and SVM, we can learn from each other and optimize the classification effect. (4) mastering the current frontier technology of VR. In the VIVE platform of HTC, the application scene of gesture recognition is constructed by using Unity game development engine, and the function module of data transmission and return result of related server is realized by Python language. Finally, the application of gesture recognition is realized.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18;TP391.9
【相似文献】
相关期刊论文 前10条
1 武霞;张崎;许艳旭;;手势识别研究发展现状综述[J];电子科技;2013年06期
2 ;新型手势识别技术可隔着口袋操作手机[J];电脑编程技巧与维护;2014年07期
3 任海兵,祝远新,徐光，
本文编号：2315826
本文链接：https://www.wllwen.com/kejilunwen/zidonghuakongzhilunwen/2315826.html
