基于智能手机的活动识别和身份识别技术研究

发布时间：2018-05-24 05:39

本文选题：智能手机 + 无监督特征学习　；参考：《国防科学技术大学》2014年硕士论文

【摘要】：随着内置了多种传感器的智能手机的广泛普及和使用,智能手机已经成为一个集通信、计算以及感知于一体的移动计算平台。利用智能手机内置的传感器如加速度传感器和陀螺仪传感器可以获取内容丰富的传感数据,如何基于加速度传感器和陀螺仪传感器进行人的日常活动识别和身份识别是目前移动感知领域研究的热点和难点问题。论文针对基于智能手机惯性传感器数据的活动识别和身份识别问题展开研究,在对相关技术研究基础上,基于无监督特征学习方法,提出了一种惯性传感器数据特征提取方法;在此基础上,将这种特征提取方法应用到活动识别和身识别当中进行实验验证。论文的主要研究内容和贡献如下:(1)针对现有的特征提取方法依赖于领域知识以及可能造成重要信息的损失等缺点,结合惯性传感器数据的内在特性,提出了一种基于无监督特征学习技术的惯性传感器数据特征提取方法,该方法首先使用无监督特征学习方法学习多个特征映射,而后再将所有特征映射拼接起来形成最终的特征计算方法,其特点是能自动地从数据中学习特征的表示方法,从而避免了手工设计的特征的一些缺点,并且从理论和实验上证验证了该方法的有效性。(2)为了评估所提出的特征提取方法在活动识别中的优劣,本文基于三种常用无监督模型提取了三种特征,使用C4.5算法、朴素贝叶斯方法和支持向量机SVM算法进行活动识别,并与其他的常用特征进行比较。实验结果表明,本文所提出的特征提取方法所提出的特征在准确率上要高于其他特征。(3)通过对基于智能手机的身份识别的关键技术进行研究,提出了一种用于身份识别的分类算法,并在上述特征提取方法提取的特征的基础上进行实验验证。实验结果表明,该算法通过选择合理的参数,能够很好地保证身份识别的可靠性和可用性。
[Abstract]:With the wide popularity and use of smart phones with multiple sensors, smart phones have become a mobile computing platform that integrates communication, computing and perception. Smart phone built-in sensors such as acceleration sensors and gyroscope sensors can be used to obtain rich sensor data, How to identify and identify people's daily activities based on acceleration sensors and gyroscopes is a hot and difficult issue in the field of mobile sensing. In this paper, the research on activity recognition and identity recognition based on smart phone inertial sensor data is carried out. Based on the research of related technology and the unsupervised feature learning method, a feature extraction method for inertial sensor data is proposed. On this basis, the feature extraction method is applied to activity recognition and body recognition for experimental verification. The main research contents and contributions of this paper are as follows: (1) aiming at the shortcomings of existing feature extraction methods, such as dependence on domain knowledge and possible loss of important information, combined with the inherent characteristics of inertial sensor data, A feature extraction method for inertial sensor data based on unsupervised feature learning technique is proposed. Firstly, unsupervised feature learning method is used to learn multiple feature maps. Then all feature maps are stitched together to form the final feature calculation method, which is characterized by the ability to automatically learn the representation of features from the data, thus avoiding some disadvantages of hand-designed features. In order to evaluate the advantages and disadvantages of the proposed feature extraction method in activity recognition, this paper extracts three features based on three common unsupervised models and uses C4.5 algorithm. Naive Bayes method and support vector machine (SVM) algorithm are used to recognize the activity and compare with other common features. The experimental results show that the feature extraction method proposed in this paper is more accurate than other features. A classification algorithm for identity recognition is proposed, and experimental verification is carried out on the basis of the features extracted by the above feature extraction methods. Experimental results show that the algorithm can ensure the reliability and availability of identification by selecting reasonable parameters.
【学位授予单位】：国防科学技术大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN929.53;TP212

【相似文献】