步态识别技术研究及系统实现

发布时间：2018-05-12 05:04

本文选题：步态识别 + OpenCV　；参考：《哈尔滨商业大学》2017年硕士论文

【摘要】：随着计算机视觉技术的快速发展,人们对安全的需求与日俱增,生物特征识别,作为对安全场合常用的身份鉴别技术,受到广泛的关注与研究。人脸、指纹等生物特征不仅要求近距离接触摄像装置,而且识别技术实现成本较高。而步态识别,具有非接触性、隐蔽性等特点,克服了传统的生物特征识别需要高分辨率图像的缺陷。本文基于OpenCV开源函数库,以步态识别在门禁系统中的应用为前提,利用C与C++(基于MFC)作为开发工具,开发实现了一套完整的步态识别系统,并深入研究步态识别中的拒识算法。本文设计一套完整的步态识别技术方案,该方案以Matlab为开发工具,对中科院自动化所提供的大型步态数据库CASIA-B中的样本进行测试,试验结果表明,该步态识别技术方案达到了理想的识别率。在此基础上,完成以下三个方面的工作:第一、利用C与C++实现整个步态识别技术,形成可以独立使用的各个功能模块;第二、搭建完整的步态图像采集环境。该图像采集环境由摄像头模块、单色背景和灯光组成,形成一套固定的采集环境,减少外部因素产生的噪声;第三、提出一种背景检测与去除算法及拒识算法。添加摄像头模块后,由于实际环境的噪声干扰,在图像预处理阶段会产生大量的噪声(阴影),不利于特征提取,将图像转换到YCrCb颜色空间中,将目标图片与背景图片的Y、Cr、Cb通道求差,通过设定的帧差阈值找到阴影位置并实现阴影去除。另外,由于门禁系统在实际应用中不仅需要满足较高的识别率,还需有拒识功能,即不仅能准确识别训练样本库中的真实样本,还能拒绝识别不属于训练样本类的虚假样本。因此,本文对步态识别中拒识模块展开研究,提出一种基于极值寻优的拒识算法。在只有训练样本的前提下,找到几何空间中所有样本投影后离散度最小的投影方向,确定拒识区间,当非训练样本的测试样本在投影方向上投影后,落在拒识区间,实现拒识功能。实验结果表明,该拒识算法能实现较高的拒识率,证实了该算法的有效性本文最终实现了一套基于OpenCV开发完整且可实际使用的步态识别系统,不仅实时性强,而且用户界面友好。
[Abstract]:With the rapid development of computer vision technology, the demand for security is increasing day by day. Biometric identification, as a commonly used identification technology in security situations, has received extensive attention and research. Face, fingerprint and other biological features not only require close contact camera, but also high cost of recognition technology. Gait recognition has the characteristics of non-contact and concealment, which overcomes the defect of traditional biometric recognition which requires high resolution image. Based on OpenCV open source function library and the application of gait recognition in access control system, a complete gait recognition system is developed by using C and C as development tools. Furthermore, the rejection algorithm in gait recognition is deeply studied. In this paper, a complete gait recognition scheme is designed. With Matlab as the development tool, the samples in the large gait database CASIA-B provided by the Chinese Academy of Sciences automation are tested. The experimental results show that, The gait recognition scheme achieves an ideal recognition rate. On this basis, the following three aspects of the work completed: first, the use of C and C to achieve the entire gait recognition technology to form independent use of each functional module; second, build a complete gait image acquisition environment. The image acquisition environment is composed of camera module, monochromatic background and light to form a set of fixed acquisition environment to reduce the noise generated by external factors. Thirdly, a background detection and removal algorithm and rejection algorithm are proposed. After adding camera module, because of the actual environment noise interference, in the image preprocessing stage will produce a large number of noise (shadow, is not conducive to feature extraction, the image will be converted to YCrCb color space, The error between the target image and the background image is obtained, and the shadow position is found through the frame difference threshold and the shadow removal is realized. In addition, the access control system not only needs to satisfy the higher recognition rate in the practical application, but also has the function of refusing recognition, that is, it can not only accurately identify the real samples in the training sample database, but also refuse to recognize the false samples that do not belong to the training sample class. Therefore, in this paper, the recognition module in gait recognition is studied, and a rejection algorithm based on extremum optimization is proposed. On the premise of only training samples, the projection direction of the minimum dispersion of all samples after projection in geometric space is found, and the rejection interval is determined. When the untrained test sample is projected on the projection direction, it falls into the rejection interval. The function of refusing recognition is realized. Experimental results show that the rejection algorithm can achieve a high rejection rate. The validity of the algorithm is verified. Finally, a complete gait recognition system based on OpenCV is developed and can be used in practice. And the user interface is friendly.
【学位授予单位】：哈尔滨商业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】