基于加速度传感器的人体摔倒检测系统设计

发布时间：2018-08-08 15:38

【摘要】：科学技术和社会经济的飞速发展,使得人们的生活品质有了飞跃,但也带来了负面的产物,人口老龄化就是其中一个。它不仅导致社会劳动力大幅缩减,劳动比例失调,同时占用了更多的社会资源用于助老和养老,加重了社会、政府和家庭的经济负担。目前我国还处于社会主义初级阶段,未富先老的社会现状使得负担人口老龄化的社会重担成为一个巨大的挑战。据统计,摔倒是引发60周岁及以上老年人非病理住院治疗和死亡的首要原因,并且这个比例每年仍在呈上升趋势。目前国内外市场,缺乏对老年人摔倒进行防护、预测和报警的成熟产品。对老年人摔倒行为和其它相似过程的误判现象尤为明显,该领域的研究仍有许多关键的科学和技术问题亟待解决。本文旨在研究基于加速度传感器的人体摔倒检测系统,包括硬件装置和检测算法两大部分,从时域和频域两个方向对摔倒行为进行了分析。论文主要工作如下:(1)从摔倒检测的研究背景和意义入手,介绍了当前的社会人口现状和摔倒的危害性。结合国内外的主流研究方法和现状,提出基于加速度传感器的人体摔倒检测系统,并指明了其便携性和实时性的特点。(2)分析了摔倒过程中人体姿态与运动加速度的变化和特征,对表征摔倒的人体部位进行了分析与选择。选取STM8L为系统微控制器,MPU6050为采集数据的传感器,HC-05为通信模块,使用VS2010搭建上位机,构建本课题系统的总体架构。通过实验建立包含日常行为、摔倒行为和类摔倒行为的数据集。(3)在时域上对人体各行为进行特征提取和分析。基于合加速度值的变化幅度和重力方向倾角值,提出了一种时域上的摔倒行为检测算法。经实验证明,该算法在区分人体动静和躺下姿态时,具有较高的识别效果。(4)在变换域,对人体各行为信息进行分析。使用快速傅里叶变换和离散余弦变换两种经典算法分别对信号进行变换,并进行对比。对预处理后的结果做特征提取和子空间特征降维。基于统计学习理论和支持向量机,采用最优分类界面作为摔倒检测中的阈值信息,提出了一种基于频域分析的摔倒行为检测算法。(5)在样本空间内,对各算法进行了验证。本文提出的基于快速傅里叶变换按二维主元分析的检测方法,取得了高达95%的识别率,证明了本文方法的优越性。最后,对系统存在的不足进行了总结,并为进一步的工作提出了意见。
[Abstract]:With the rapid development of science and technology and social economy, the quality of life of people has made a leap, but also brought negative products, the aging population is one of them. It not only leads to the sharp reduction of social labor force, labor ratio imbalance, but also occupies more social resources for the elderly and old-age care, and increases the economic burden of the society, the government and the family. At present, our country is still in the primary stage of socialism, and the present social situation makes it a great challenge to bear the burden of aging population. According to statistics, fall is the leading cause of non-pathological hospitalization and death in the elderly aged 60 and above, and the rate is still rising every year. At present, domestic and foreign markets, the lack of elderly fall protection, prediction and alarm of mature products. The misjudgment of fall behavior and other similar processes in the elderly is particularly obvious. There are still many key scientific and technical problems to be solved in this field. The purpose of this paper is to study the human fall detection system based on acceleration sensor, including hardware device and detection algorithm, and analyze the fall behavior in time domain and frequency domain. The main work of this paper is as follows: (1) starting with the research background and significance of fall detection, this paper introduces the current situation of social population and the harm of fall. Combined with the main research methods and present situation at home and abroad, a human fall detection system based on acceleration sensor is proposed, and the portable and real-time characteristics of the system are pointed out. (2) the changes and characteristics of human posture and motion acceleration during fall are analyzed. The analysis and selection of the human body parts characterized by the fall were carried out. STM8L is selected as the microcontroller of the system MPU6050 as the sensor of collecting data, HC-05 as the communication module, and the host computer is built with VS2010, and the overall structure of the system is constructed. The data sets including daily behavior, fall behavior and similar fall behavior are established by experiments. (3) feature extraction and analysis of human behavior in time domain. Based on the amplitude of the combined acceleration and the inclination of the gravity direction, an algorithm for detecting the fall behavior in time domain is proposed. The experimental results show that the algorithm has a high recognition effect in distinguishing human motion from lying down posture. (4) in the transform domain, the human behavior information is analyzed. Two classical algorithms, Fast Fourier transform (FFT) and discrete CoSine transform (DCT), are used to transform the signal. After pretreatment, feature extraction and subspace feature reduction are done. Based on the statistical learning theory and support vector machine, the optimal classification interface is used as the threshold information in fall detection, and a fall behavior detection algorithm based on frequency domain analysis is proposed. (5) in the sample space, the algorithms are verified. The detection method based on fast Fourier transform (FFT) based on two-dimensional principal component analysis (PCA) is presented in this paper, and the recognition rate is up to 95%, which proves the superiority of this method. Finally, the shortcomings of the system are summarized, and suggestions for further work are put forward.
【学位授予单位】：重庆大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP212.9;TP18

【相似文献】