基于决策树的低功耗手机活动识别系统的设计与实现

发布时间：2018-06-11 12:03

本文选题：决策树 + 活动识别　；参考：《南昌大学》2016年硕士论文

【摘要】：活动识别是健康监测领域的一个重要环节,使用智能手机进行人体的活动识别具有普及率高、便于携带等优势。但是目前的手机活动识别研究中仍存在识别准确度不高、功耗太大等问题,以及在监测过程中必须将手机固定于身体的某个部位的局限性。为了解决上述问题,本论文首先引入欧氏距离作为分类特征值,采用决策树作为分类算法,有效解决坐和站立两种静态活动较难识别的问题,从而提高活动识别的准确率。对6名志愿者采集的活动数据进行实验验证,平均识别准确率达92.63%。为了解决活动识别应用过程中引起的过高功耗问题,本论文引入了工作周期概念对加速度传感器的数据采集周期进行动态管理,有效降低系统对手机电量的消耗,实现活动识别应用的低功耗。为了解决在活动识别过程中,用户必须将手机固定置放于身体某个部位的局限性,本论文使用了与方向无关的量信号幅度作为分类特征值,同时使用一阶回归系数和四分位距作为特征量,从而使系统可检测随意置放手机的活动,实验证明了该方法的有效性。最后,论文采用Java语言,使用Eclipse开发工具,基于Android开发平台,设计与实现了一个能实际使用的低功耗手机活动识别系统。
[Abstract]:Activity recognition is an important link in the field of health monitoring. The use of smart phones for human activity recognition has the advantages of high prevalence and easy to carry. However, there are still some problems in mobile activity recognition, such as low recognition accuracy, too much power consumption, and the limitation that the mobile phone must be fixed to a certain part of the body during the monitoring process. In order to solve the above problems, Euclidean distance is introduced as the classification eigenvalue and decision tree is used as the classification algorithm to effectively solve the problem that it is difficult to identify two kinds of static activities: sitting and standing, so as to improve the accuracy of activity recognition. The activity data collected by 6 volunteers were tested and the average recognition accuracy was 92.63. In order to solve the problem of excessive power consumption caused by the application of activity identification, this paper introduces the concept of work cycle to dynamically manage the data acquisition cycle of accelerometer, which can effectively reduce the consumption of mobile phone power. The low power consumption of the activity recognition application is realized. In order to solve the problem that the mobile phone must be fixed in a certain part of the body in the process of activity recognition, this paper uses the amplitude of the signal which is independent of the direction as the characteristic value of the classification. The first order regression coefficient and the quartile distance are used as the characteristic variables at the same time, so that the system can detect the mobile phone activity at random. The experimental results show that the proposed method is effective. Finally, a low power mobile activity recognition system is designed and implemented based on Android development platform using Java language and Eclipse development tools.
【学位授予单位】：南昌大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41;TP212.9

【相似文献】