光学 精密工程, 2017, 25 (1): 182, 网络出版: 2017-03-10
基于支持向量机的跌倒检测算法研究
Research on fall detection system based on support vector machine
跌倒检测 惯性传感器 机器学习 支持向量机 粒子群优化 径向基函数 fall detection inertial sensor machine learning SVM Particle Swarm Optimization RBF
摘要
实时跌倒检测能有效降低老人因跌倒导致的身心伤害, 提高老人的独居能力和健康水平。为提高基于惯性传感器的跌倒检测系统的准确率, 降低系统误报率和漏报率, 提出了应用基于径向基函数的支持向量机算法实现跌倒判定。首先, 应用佩戴在人体腰间的便携式跌倒检测系统完成数据的采集; 然后, 利用基于径向基函数(RBF)的SVM分类器标记疑似跌倒行为, 并利用粒子群算法完成分类算法中惩罚因子C和RBF参数g的优化。结果表明, 在区分跌倒与类似跌倒的日常活动时, 基于SVM算法的跌倒检测系统准确率、误报率和漏报率分别为97.67%, 4.0%和0.67%。与传统的阈值方法相比, 跌倒检测性能有很大提高, 从而加强了该系统在老人跌倒检测中的应用。
Abstract
Real-time fall detection has great advantages of reducing physical and psychological damage in senior citizens group after falls and improving solitude ability and health level of senior citizens. A support vector machine (SVM) algorithm, which is based on RBF(Radial Basis Function) and applied to achieve fall detection, has been proposed in order to improve accuracy rate and lower false positive and false negative rate of fall detection system on the basis of inertial sensor. First, the system completes data collection by portable inertial sensing system at waist; then, it utilizes RBF-based SVM classifier to identify suspected fall behaviors and Particle Swarm Optimization to complete optimization of penalty factor ‘C’ and RBF argument ‘g’ in sorting algorithm. The falls and similar falls daily activities distinguishing experimetal results indicate that accuracy rate, false positive and false negative rate based on SVM algorithm are 9767%, 4.0% and 0.67% respectively. Compared with traditional threshold methods, the performance of proposed method on fall detection is promoted remarkably, so it can conclude that the appliance of the system in senior citizens fall detection is enhanced as well.
裴利然, 姜萍萍, 颜国正. 基于支持向量机的跌倒检测算法研究[J]. 光学 精密工程, 2017, 25(1): 182. PEI Li-ran, JIANG Ping-ping, YAN Guo-zheng. Research on fall detection system based on support vector machine[J]. Optics and Precision Engineering, 2017, 25(1): 182.