基于KNN算法的跑步姿态监测与识别

发布时间：2018-04-23 02:35

本文选题：跑步姿态识别 + 运动跟踪　；参考：《东华大学》2017年硕士论文

【摘要】：随着生活水平的不断提高,借助于跑步机的健身跑步成为目前的有氧健身运动方式之一,跑步对颈椎、脊椎、心脏等各个人体机能都有好处,但是错误的跑步姿势也会对运动对象造成身体关节的损伤以及其他危害。为了能让更多的健身运动者以正确的姿势跑步,本文针对跑步机上的跑步视频进行了研究。随着科学技术以及信息技术的持续进步,视频监控在各种领域中的应用已经得到了普及,对动态目标的跟踪检测技术是应用视频进行监控的基础和关键,该项技术将图像处理、模式识别以及AI人工智能等多个领域融合到一起。在机器视觉领域中,以“人体运动”为对象的研究一直在不断进行,从人体检测、位置跟踪、运动轨迹跟踪,到现在姿态识别、人体动作识别,已经有了一系列相关的研究算法和成果。本文以跑步机上的跑步视频作为实验对象,主要进行了以下这些工作:(1)首先对当前在人体运动识别技术的应用领域的背景以及意义这两个部分的内容进行了简单论述,并结合本文中的运动对象,对人体运动识别领域较为完整的研究成果进行比较,同时对本文的研究中会出现的难点进行了分析,为本文进行研究提供了基本思路。(2)对于运动视频中的对象跟踪,为了获取更准确的运动轨迹,将基于Kinovea软件的路径跟踪以及基于MeanShift算法的位置跟踪方法进行了对比,选取跟踪结果更为准确且能提高分类算法运行的跟踪方法,本文选用基于Kinovea软件的跟踪方法进行路径跟踪。(3)介绍了当前常用的识别算法。在进行分类器的设计时,结合研究对象的运动模型特点,选用适合本文研究对象的分类器,在KNN分类器的基础上,降低样本维度进行测试,提高分类识别效果。(4)进行试验测试,邀请实验人员进行跑步测试,对其跑步姿势视频进行分析,分别从背面和侧面两个角度分别进行分类识别,再获取结论之后对其跑步姿势提出校正训练建议。论文在最后总结了本文全部的研究工作,同时对接下来的研究目标和方法进行了分析和展望。
[Abstract]:With the continuous improvement of living standards, running with treadmill fitness has become one of the current aerobic exercise methods, running is good for the cervical spine, spine, heart and other human functions. But the wrong running posture can also cause joint damage and other hazards. In order to make more fitness athletes run in the right position, this paper studies the running video on treadmill. With the continuous progress of science and technology and information technology, the application of video surveillance in various fields has been popularized. The technology of tracking and detecting dynamic targets is the basis and key of video surveillance. Multiple fields such as pattern recognition and AI are combined. In the field of machine vision, the research on "human motion" has been going on all the time, from human body detection, position tracking, motion track tracking, to posture recognition, human motion recognition, There have been a series of related algorithms and results. This article takes the running video on the treadmill as the experimental object, mainly carries on the following work: 1) first of all, has carried on the brief discussion to the current application background and the significance of the two parts in the human body motion recognition technology domain. Combined with the moving object in this paper, the relatively complete research results in the field of human motion recognition are compared. At the same time, the difficulties that will appear in the research are analyzed. In order to obtain more accurate motion trajectory, the path tracking method based on Kinovea software and the position tracking method based on MeanShift algorithm are compared. The tracking method which is more accurate and can improve the running of the classification algorithm is selected. In this paper, the path tracking method based on Kinovea software is selected to introduce the commonly used recognition algorithms. In the design of classifier, combining the characteristics of motion model of the research object, selecting the classifier suitable for the research object in this paper, on the basis of KNN classifier, reducing the sample dimension to test, improving the classification recognition effect. The participants were invited to conduct the running test, the video of their running posture was analyzed, and the running posture was classified and recognized from the two angles of the back and the side respectively, and then the conclusion was obtained, and then the training suggestions were put forward for their running posture correction. At the end of this paper, all the research work is summarized, and the following research objectives and methods are analyzed and prospected.
【学位授予单位】：东华大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】