智能视频监控系统中若干生物特征识别研究

发布时间：2018-02-21 23:51

  本文关键词： 智能视频监控 人脸识别 步态识别 体态识别 骨骼跟踪　出处：《电子科技大学》2014年硕士论文　论文类型：学位论文

【摘要】：智能视频监控系统近几年在计算机视觉领域发展迅速,并成为该领域的重要研究方向。生物特征识别技术是指利用人体的固有特征进行身份鉴别的计算机技术,广泛应用于银行、司法等身份鉴别领域。将人体生物特征识别技术应用到智能视频监控系统中,构建自动身份识别及处理系统。应用于智能视频监控中的生物特征识别受光照、环境噪声以及人体运动等因素的干扰,实现起来较基于静态图片的生物特征识别困难。因此,分析、研究智能视频监控系统中对生物特征识别造成干扰的因素具有很大的应用价值。研究智能视频监控系统中的生物特征识别方法是本文的研究重点。本文研究目的是通过对智能视频监控系统中生物特征识别算法进行研究,提升系统中人体身份识别的准确率,为智能视频监控系统的真正应用做出贡献。在监控视频中,生物特征识别具有非接触性和非侵犯性的特点。其中,能够用于身份鉴别的特征有人脸特征、步态特征以及体形体态特征。本文分别对这三种特征进行分析研究,并提出相应的特征识别算法。本文的主要工作有:1、研究人脸识别技术,首先对智能视频监控中获取的人脸图像处理方法进行分析,通过预处理得到利于识别的人脸图像。然后通过对传统的LBP算法进行研究和改进,提出了改进后的LBP人脸识别算法。该方法通过计算区域内的均值和方差,求得该邻域的四值模式。通过实验验证,该算法的识别率较传统LBP算法有所提高、鲁棒特性好。最后将改进的LBP算子应用于人脸识别系统。2、研究步态特征提取方法,提出基于Kinect的步态特征提取方法。运用Kinect能够提取三种步态特征,分别是:双腿关节点角度信息、行走时的步幅特征以及三维人体轮廓描述子。介绍了Kinect的深度获取以及骨骼获取原理,并通过Kinect的坐标空间转换得到三维人体轮廓。最后采用最近邻分类器和k-近邻分类器进行实验,实验表明文中提出的基于Kinect的步态识别方法有效,识别率达到84%。3、尝试性的提出体态识别方法,分析体态特征用于身份鉴别的理论依据以及限制条件。定义人体体态特征,然后使用Kinect的骨骼跟踪功能对人体体态特征进行提取。最后分别使用标准欧式距离分类器、方差倒数加权欧式距离分类器以及决策树分类方法进行实验,表明该人体体态识别方法有效,最高识别率达到87%。
[Abstract]:Intelligent video surveillance system has developed rapidly in the field of computer vision in recent years, and has become an important research direction in this field. It is widely used in bank, judicial and other identification fields. The technology of human biometric identification is applied to the intelligent video surveillance system, and the automatic identification and processing system is constructed. The biometric identification used in intelligent video surveillance is illuminated. It is more difficult to realize biometric recognition based on static picture because of the interference of environmental noise and human body movement. It is of great value to study the factors that interfere with biometric identification in intelligent video surveillance system. The research of biometric recognition method in intelligent video surveillance system is the focus of this paper. It is based on the research of biometric recognition algorithm in intelligent video surveillance system. Enhance the accuracy of human identification in the system, and contribute to the real application of intelligent video surveillance system. In surveillance video, biometric recognition has the characteristics of non-contact and non-invasive. The features that can be used for identification are facial features, gait features and body features. The main work of this paper is to study the face recognition technology. Firstly, the face image processing method obtained from intelligent video surveillance is analyzed. Through the research and improvement of the traditional LBP algorithm, an improved LBP face recognition algorithm is proposed, which calculates the mean value and variance in the region. The quad-valued pattern of the neighborhood is obtained. The experimental results show that the recognition rate of the algorithm is better than that of the traditional LBP algorithm. Finally, the improved LBP operator is applied to face recognition system .2and the gait feature extraction method is studied. A gait feature extraction method based on Kinect is proposed. Three gait features can be extracted by using Kinect. The characteristics of walking stride and 3D human outline descriptor. The principle of depth acquisition and bone acquisition of Kinect is introduced. Finally, the nearest neighbor classifier and k- nearest neighbor classifier are used to experiment. The experiment shows that the proposed gait recognition method based on Kinect is effective. The recognition rate is 84%. 3. Try to put forward a method of body recognition, analyze the theoretical basis and limiting conditions of body feature used for identity identification, and define the body characteristics of human body. Then we use the skeleton tracking function of Kinect to extract the human body features. Finally, we use standard Euclidean distance classifier, variance-reciprocal weighted Euclidean distance classifier and decision tree classification method to carry out experiments. The results show that the method is effective and the highest recognition rate is 87.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.41;TN948.6
，

本文编号：1523159