基于多特征的人体行为识别的研究

发布时间：2018-03-28 19:40

本文选题：无线传感器网络　切入点：行为识别　出处：《南京邮电大学》2017年硕士论文

【摘要】：随着无线通信技术以及传感器制造技术的快速发展,无线传感器网络应用领域涉及生活休闲、医疗保健、军事通信、应急救灾及航空航天等方面。目前,无线传感器网络中的人体行为识别具有便携性好、成本低、抗干扰等优点,受到了人们的广泛关注。人体行为的识别方法种类很多,从数据采集方式到识别分类算法,行为识别系统中的各个环节都具有不同的特点。本文主要针对特征提取、识别分类算法进行了研究,并在公开的人体动作识别数据库上进行了实验,对实验结果进行了评估与分析。本文主要工作如下:(1)分析了基于无线传感器网络的人体行为识别课题的研究背景和意义,对基于三轴加速度传感器的人体行为识别的信号采集方式、数据预处理方法、特征提取等几个模块进行了概述和分析。(2)分析比较了SVM、KNN、NB等算法用于行为识别时的优缺点。考虑到KNN算法模型的优势,本文通过Relief F算法计算特征的权重,改进了KNN算法,通过实验分析了其改进后的识别准确率。(3)将邻近类作为局部基,在稀疏表示的人体行为识别算法中引入KNN算法的思想,给出了一种新的稀疏近邻表示分类方法。实验结果表明了这种方法可以有效弥补原算法存在的缺点,降低复杂度且得到了很好的识别准确率。(4)采用一种基于块稀疏模型的人体行为识别方法,充分利用了人体行为模型内在块稀疏结构,将人体行为识别问题转化为稀疏表示问题。通过块稀疏贝叶斯学习算法,求解待测样本的稀疏系数并进行分类,实验结果说明了该种方法能有效提高人体行为识别率。
[Abstract]:With the rapid development of wireless communication technology and sensor manufacturing technology, wireless sensor network applications are related to life and leisure, medical care, military communications, emergency relief and aerospace. Human behavior recognition in wireless sensor networks has many advantages, such as good portability, low cost, anti-jamming and so on. There are many kinds of human behavior recognition methods, from data acquisition to recognition and classification algorithm. Each link in the behavior recognition system has different characteristics. This paper mainly focuses on the feature extraction, recognition and classification algorithm, and carries on the experiment on the open human body motion recognition database. The main work of this paper is as follows: 1) the background and significance of the research on human behavior recognition based on wireless sensor networks are analyzed. The signal acquisition method and data preprocessing method of human body behavior recognition based on triaxial acceleration sensor are discussed. Several modules such as feature extraction are summarized and analyzed. (2) the advantages and disadvantages of SVMKNNNNB and other algorithms in behavior recognition are compared. Considering the advantages of KNN algorithm model, this paper improves the KNN algorithm by calculating the weight of features by Relief F algorithm. The improved recognition accuracy is analyzed experimentally. (3) the neighborhood class is used as the local basis, and the idea of KNN algorithm is introduced into the sparse representation of human behavior recognition algorithm. A new sparse nearest neighbor representation classification method is presented. The experimental results show that this method can effectively compensate for the shortcomings of the original algorithm. A method of human behavior recognition based on block sparse model is proposed, which makes full use of the block sparse structure of human behavior model. The problem of human behavior recognition is transformed into a sparse representation problem. The sparse coefficients of samples under test are solved and classified by block sparse Bayesian learning algorithm. The experimental results show that this method can effectively improve the recognition rate of human behavior.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP212.9;TN929.5

【参考文献】