基于多传感器的人体运动识别算法与应用研究

发布时间：2017-12-31 09:16

本文关键词：基于多传感器的人体运动识别算法与应用研究　出处：《重庆邮电大学》2016年硕士论文　论文类型：学位论文

【摘要】：随着社会老龄人口数量的增加和家庭结构的变化,中国“空巢”老人数量大幅度增加,老人的生活从以前依靠子女照顾到现在自动求助模式。因此老年人日常自动监测会对老年人的日常生活以及社会的稳定都会有很大的帮助。针对现有的基于多传感器的人体运动识别算法复杂度较高很难将其搬移到便携的终端以及系统的能耗大需要频繁更换系统电池或充电的问题,本文提出了基于两层决策树识别器的人体运动识别算法,并以两层决策树算法为基础,设计了一套完整的跌倒检测系统,实现了对老年人的跌倒进行识别。本文的主要工作如下:1.分析和总结了现有的基于传感器的人体运动识别的理论和研究方法,并对数据采集模块、数据预处理的方法、特征提取和选择的技术、分类识别的识别器以及无线短距离通信技术等五个方面进行了详细分析。2.根据日常生活中人体运动的特点以及采集的数据特性,提出了一种基于两层决策树识别器的人体运动识别算法,通过减少对加速度数据的提取和处理来降低系统的能耗和算法的复杂度。在本研究中还采用了卡尔曼滤波对数据进行滤波处理以及长度为2s的半重叠的固定窗口对数据进行加窗处理,来提高系统的识别率以及降低系统算法的复杂度。3.使用MPU6050传感器模块和蓝牙通信模块采集20位实验者在不同状态下八种运动的数据用于训练和验证识别器的准确率。用WEKA软件验证的结果显示,基于两层决策树识别器的准确率分别为98.44%和94.16%,并对比了其他分类识别器,其各方面的性能都是较优的。4.设计并实现了一套完整的老年人跌倒检测报警系统,具体跌倒检测算法是基于两层决策树识别器的。并通过实验数据表明,本文设计的跌倒检测系统的准确识别率达到了95%,是有一定市场利用价值的。
[Abstract]:With the increase in the number of elderly people in society and the changes in family structure, the number of "empty nests" in China has increased by a large margin. The life of the old people depends on their children to take care of them now. Therefore, automatic daily monitoring of the elderly will be of great help to the daily life of the elderly as well as the stability of the society. It is very difficult to move the sensor to portable terminals with high complexity and the problem that the energy consumption of the system needs to replace the battery or charge the system frequently. In this paper, a human motion recognition algorithm based on two-layer decision tree recognizer is proposed. Based on the two-layer decision tree algorithm, a complete fall detection system is designed. The main work of this paper is as follows: 1. The existing theory and research methods of human motion recognition based on sensor are analyzed and summarized, and the data acquisition module is given. Methods of data preprocessing, feature extraction and selection techniques. The classifier and wireless short distance communication technology are analyzed in detail. 2. According to the characteristics of human body movement and data collection in daily life. A human motion recognition algorithm based on two-layer decision tree recognizer is proposed. The energy consumption and algorithm complexity of the system are reduced by reducing the extraction and processing of acceleration data. In this study, Kalman filter is also used to filter the data and a semi-overlapping fixed window of 2 s in length is used in this study. The port carries on the window processing to the data. To improve the system recognition rate and reduce the complexity of the system algorithm .3. using the MPU6050 sensor module and Bluetooth communication module to collect 20 experimenters in different states of eight kinds of motion data for training. And verify the accuracy of the recognizer. The results of the verification with WEKA software show. The accuracy of the two-layer decision tree recognizer is 98.44% and 94.16, respectively, and the other classifiers are compared. The performance of each aspect is better. 4. A set of integrated fall detection and alarm system for the elderly is designed and implemented. The specific fall detection algorithm is based on a two-layer decision tree recognizer. The accurate recognition rate of the fall detection system is 95%, which has certain market value.
【学位授予单位】：重庆邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP212.9

【参考文献】