基于穿戴式的人体跌倒检测系统研究与设计

发布时间：2018-08-02 09:58

【摘要】：人体姿态在很多领域发挥着重要的作用,比如一些病人的康复训练、人机交互游戏、电影中一些特效等。如何快速而准确的检测与识别出人体某种姿态已成为当前一个热门的研究课题。目前人体姿态检测方案大体可以分为两类:一是通过检测人体周围环境信息如人体视频图像、声音、电磁波以及红外线等信息的变化作为判定依据进行相应姿态的判定;二是通过人体佩戴的一些装置如三轴加速度传感器、压力传感器和惯性传感器等采集人体姿态相关的一些数据来推断与识别人体姿态。结合当前我国空巢老人多,加上跌倒事故的频繁发生,并且跌倒后有时得不到及时救助的实际情况。本系统从携带方便、低功耗、实时性好三个方面考虑,设计了仅需一个三轴加速度传感器的人体姿态检测系统,该装置适合佩戴于人体腰部位置。此外,系统设计自动求助功能和主动求助功能。可以在老人发生跌倒后得到及时救助。检测系统由信号采集模块、处理模块以及通讯模块组成。传感器模块负责采集加速度数据,并进行预处理;处理模块提取合加速度SMVA、合加速度微分绝对平均值M4DS、加速度信号强度SM4、人体倾斜角BT4等相关特征值;通讯模块负责对人体进行定位和求助短信发送。论文中根据检测到的数据分别采用当前常用的阈值法、本文改进后的阈值法(多级跌倒检测算法)和支持向量机的奇异值分解算法进行了跌倒检测实验,对三种算法的跌倒检测判断的准确率的比较,改进后的多级姿态检测算法判断跌倒检测的准确率比一般阈值法的要高,但与支持向量机的检测算法的准确率差距不大。但是多级跌倒检测算法实时性较支持向量机算法好。综合实时性和准确率,决定选用多级跌倒检测算法作为最终方案。在此基础上完成了跌倒检测系统的软硬件设计。通过对系统各个功能的测试,能够很好的实现人体跌倒检测和自动发出求救短信功能。检测效果的准确率和实时性均达到预期目的。
[Abstract]:Human posture plays an important role in many fields, such as rehabilitation training for patients, interactive games, special effects in movies and so on. How to detect and recognize human posture quickly and accurately has become a hot research topic. At present, human attitude detection schemes can be divided into two categories: one is the determination of posture by detecting the changes of human body environment information such as human body video image, sound, electromagnetic wave, infrared ray and so on; Secondly, some devices such as triaxial acceleration sensor, pressure sensor and inertial sensor are used to infer and recognize human posture. Combined with the fact that there are many empty nest elderly people in our country, and the frequent fall accidents occur, and sometimes can not get timely rescue after falling. Considering the convenience of carrying, low power consumption and good real-time performance, the system designed a three-axis acceleration sensor for human body attitude detection system, which is suitable for the waist position of the human body. In addition, the system designed automatic help function and active help function. You can get help in time after an old man falls down. The detection system consists of signal acquisition module, processing module and communication module. The sensor module is responsible for collecting acceleration data and preprocessing, processing module extracts the acceleration SMVA, the differential absolute average of acceleration M4DS, the acceleration signal intensity SM4, the human body tilt angle BT4 and other related characteristic values. The communication module is responsible for locating the human body and sending short messages for help. According to the detected data, the current commonly used threshold method is used in this paper. The improved threshold method (multilevel fall detection algorithm) and the singular value decomposition algorithm of support vector machine (SVM) are used in the fall detection experiment. Compared with the accuracy of the three algorithms, the improved multi-level attitude detection algorithm has higher accuracy than the general threshold method, but the accuracy of the improved multi-level attitude detection algorithm is not far from that of the support vector machine algorithm. But multilevel fall detection algorithm is better than support vector machine algorithm in real time. Considering the real-time and accuracy, it is decided to choose the multi-level fall detection algorithm as the final scheme. On this basis, the software and hardware design of the fall detection system is completed. By testing all the functions of the system, we can realize the function of human fall detection and send out the short message automatically. The accuracy and real-time of the detection effect reach the expected goal.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP212.9

【参考文献】