基于多传感融合的老人跌倒检测算法研究

发布时间：2018-09-14 12:29

【摘要】：随着老龄化的加剧,以及空巢家庭的增多,老人的意外跌倒也随之增多。而及时对跌倒的老人进行救助,可有效地降低因跌倒引起的伤残率与死亡率。因此,老人跌倒检测系统以及跌倒检测算法成为研究热点,通过现代科学技术来检测跌倒行为,尽可能地降低老年人因跌倒带来的伤害。穿戴式的跌倒检测系统具有不限制于室内、不受周边环境干扰、方便携带等特点,能较好地满足老人跌倒检测系统的需求。本文针对基于穿戴式系统中的跌倒检测算法进行相关研究,主要工作如下:1.在基于加速度与姿态角阈值的跌倒检测算法的基础上,针对误判率较高的问题,本文在生物医学领域步态分析的启发下,将足底压力与加速度、姿态角信息相结合,提出了基于多行为特征融合的跌倒检测算法。该算法首先对人体加速度与姿态角进行判定,若均超过阈值,再对足底压力进行阈值判定,构建足底压力矩阵并进行计算。若两个矩阵都为零,则判定为跌倒,其余情况均判定为不跌倒。算法中加速度、姿态角、足底压力的阈值通过粒子群算法进行确定。通过仿真实验证明,该算法对弯腰、下蹲行为的漏报率较低,对跌倒的识别正确率较高。2.针对基于阈值的跌倒检测算法对阈值的依赖性较大且个体行为差异对阈值选取影响较大的问题,提出了基于支持向量机的跌倒检测算法。该算法首先对原始获取的行为数据进行特征处理,转化为18维的行为特征向量。利用k折交叉验证选取算法中的最优参数,并用此最优参数对行为特征进行训练。用训练得到的模型对行为测试集进行类别检测,以获得该行为特征标签来识别跌倒行为。通过仿真实验,有效检测出跌倒行为,验证了该算法的可靠性。3.针对上述的检测效果更优的SFDA跌倒检测算法,设计了基于云的跌倒检测系统。该系统主要由穿戴设备与云服务器部分组成,穿戴设备的足部模块、胸部模块以及GPRS传输模块将采集的人体行为数据上传至云端,云端以LNMP(Linux+Nginx+MySQL+PHP)为基础运行环境,对上传的行为数据进行JSON解析、SFDA算法的检测与训练、数据存储等一系列后台处理以及在前端进行显示。测试实验结果表明,算法在基于云的跌倒检测系统中的可行性较好,且该系统对跌倒行为的检测有很高的准确率。
[Abstract]:With the aggravation of aging and the increasing number of empty nest families, the accidental fall of the elderly also increases. Timely rescue can effectively reduce the disability rate and mortality caused by falls. Therefore, the fall detection system and the fall detection algorithm have become the focus of research, through modern science and technology to detect fall behavior, as far as possible to reduce the elderly fall caused by harm. The wearable fall detection system can meet the needs of the elderly fall detection system because it is not limited to the indoor, free from the interference of the surrounding environment and easy to carry. In this paper, the fall detection algorithm based on wearable system is studied. The main work is as follows: 1. On the basis of the fall detection algorithm based on acceleration and attitude angle threshold, aiming at the problem of high error rate, this paper combines plantar pressure with acceleration and attitude angle information under the inspiration of gait analysis in biomedical field. A fall detection algorithm based on multi-behavior feature fusion is proposed. The algorithm firstly determines the acceleration and attitude angle of human body. If all of them exceed the threshold value, then the foot pressure threshold is determined, and the plantar pressure matrix is constructed and calculated. If both matrices are zero, it is judged to be a fall, and the rest of the cases are determined not to fall. The threshold of acceleration, attitude angle and plantar pressure is determined by particle swarm optimization (PSO). The simulation results show that the algorithm has a lower false report rate for bending and squat behavior, and a higher accuracy rate for recognition of falls. In order to solve the problem that threshold-based fall detection algorithm is dependent on threshold and individual behavior difference has great influence on threshold selection, a fall detection algorithm based on support vector machine (SVM) is proposed. The algorithm firstly processes the original behavior data and transforms them into 18 dimensional behavior feature vectors. The optimal parameters in the algorithm are verified by k-fold crossover, and the optimal parameters are used to train the behavior characteristics. The training model is used to detect the behavior test set to obtain the behavior feature tag to identify the fall behavior. Through the simulation experiment, the fall behavior is effectively detected, and the reliability of the algorithm is verified. 3. 3. A cloud based fall detection system is designed for the above SFDA fall detection algorithm. The system is mainly composed of wearable device and cloud server. The foot module, chest module and GPRS transmission module upload the collected human behavior data to the cloud. The cloud is based on LNMP (Linux Nginx MySQL PHP). The uploaded behavior data is detected and trained by JSON parsing algorithm, and a series of background processing, such as data storage, and display on the front end, are carried out. The experimental results show that the algorithm is feasible in the cloud-based fall detection system, and the system has a high accuracy in the detection of fall behavior.
【学位授予单位】：浙江理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP274;TP18

【参考文献】