基于人体几何特征的步态识别算法研究及应用平台设计

发布时间：2018-01-28 21:06

本文关键词： 智能视频监控步态识别人体几何特征关节点提取监控平台设计开发　出处：《浙江大学》2017年硕士论文　论文类型：学位论文

【摘要】：近几年全球范围内恐怖袭击和犯罪活动不断发生,各个国家对于社会公共安全越来越重视,视频监控领域因此成为近几年的研究热点。人作为视频监控中最重要的主体之一,将同一个人在不同监控摄像头中进行有效识别与追踪具有非常大的应用价值,行人重识别也因此成为了视频监控领域中非常重要的研究方向。基于生物特征的识别因其特征的唯一性成为现阶段行人重识别的研究重点,而步态特征相对于其他生物特征(比如人脸指纹等)具有可远距离观察、采集方便等优势,在视频跟踪和识别等安防场景中,步态识别将起到越来越大的作用,具有非常广阔的应用前景。论文以步态识别为研究内容,提出了一种基于人体几何特征的步态识别算法,同时设计开发了一套面向应用的智能视频监控平台。论文的主要研究工作和贡献如下:(1)在前人目标高度宽度测量的基础算法上,对消隐点的提取算法以及目标高度宽度测量的误差矫正进行改进,设计了一套基于单目视觉的目标高度宽度自动测量算法,并应用在现实场景中。实验结果显示高度测量百分误差约为1%,宽度测量百分误差约为3%,准确度较高。(2)提出了一种侧面视角下基于人体几何特征的行人重识别方法,人体几何特征包括身高、步幅宽度以及关节角度,并且将人体下半身关节点提取算法进行改进,提高关节点的定位精度,同时针对步态识别类间模糊类内差大的特点,选用了一种基于距离学习的分类方法。本文提出的算法在中科院的步态数据集clothes类别上的识别率为56.5%,超越了现有的CGI、bSHI等算法。(3)基于实验室搭建的视频监控网络设计开发了一个智能视频监控平台,实现了视频回放、视频内容筛选以及步态识别的功能。该智能视频分析平台的实现能够大大减少人工查看监控视频的时间,提升系统的存储能力,同时也使得行人重识别过程变得更加自动化。该平台采用模块化的方式进行开发,能够将实现的算法或者功能无缝地添加到平台中,并且应用到实际场景。
[Abstract]:In recent years, terrorist attacks and criminal activities have occurred all over the world, and various countries are paying more and more attention to social public safety. As one of the most important subjects in video surveillance, the identification and tracking of the same person in different surveillance cameras has great application value. Therefore, pedestrian recognition has become a very important research direction in the field of video surveillance. Because of its uniqueness, biometric recognition has become the focus of pedestrian recognition research. Gait features have the advantages of remote observation and easy acquisition compared with other biometric features such as face fingerprint. Gait recognition will play a more and more important role in security scenes such as video tracking and recognition. This paper presents a gait recognition algorithm based on geometric features of human body. At the same time, an application-oriented intelligent video surveillance platform is designed and developed. The main research work and contributions of this paper are as follows: 1) on the basis of the previous algorithm of height width measurement. By improving the algorithm of extracting hidden point and correcting the error of target height width measurement, a set of automatic height width measurement algorithm based on monocular vision is designed. The experimental results show that the error of height measurement is about 1 percent, and the error of width measurement is about 3%. A method of pedestrian recognition based on geometric features of human body is proposed in this paper, which includes height, stride width and angle of joint. And the lower body of the human body node extraction algorithm is improved to improve the location accuracy of the node, and at the same time for gait recognition of the characteristics of large intra-class fuzzy class difference. A distance learning based classification method is selected. The recognition rate of the proposed algorithm on the gait data set clothes category of the Chinese Academy of Sciences is 56.5, surpassing the existing CGI. BSHI and other algorithms. 3) based on the laboratory video surveillance network design and development of an intelligent video surveillance platform to achieve video playback. The realization of the intelligent video analysis platform can greatly reduce the time of manually viewing the monitored video and improve the storage capacity of the system. At the same time, it also makes the pedestrian recognition process more automated. The platform is developed in a modular way, which can seamlessly add the algorithm or function to the platform and apply it to the actual scene.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TN948.6

【相似文献】