基于人体姿态的P SO-SVM特征向量跌倒检测方法

在可穿戴设备检测人体跌倒情况时,单一采用加速度阈值判别方法不能完整表征人体跌倒行为变化的信息,导致对跌倒信息误判.为此,提出了一种基于人体姿态的PSO-SVM特征向量跌倒检测算法.首先通过MEMS加速度传感器节点采集人体姿态数据,并利用共轭梯度法对采集的数据进行优化处理,降低非线性误差;然后,利用支持向量机SVM(Support Vector Machine)分类器检测跌倒行为,并通过粒子群PSO(Particle Swarm Optimization)算法对SVM参数进行优化,获得最佳分类模型,根据SVM分类模型对采集的姿态数据进行分析,判断是否跌倒;最后根据人体姿态角,构建融合人体姿态角的PSO-SVM特征向量,检测跌倒过程的具体信息.实验结果表明:该检测方法取得95.5%的识别率,能够较好地区分其他非跌倒性动作,检测精度较其他方法较高,均方根误差较小,有较好的鲁棒性.

PSO-SVM Feature Vector Fall Detection Algorithm Based on Human Postures

Adopting the method of accelerating threshold can not demonstrate the variation of falling message. It will lead to the misjudgement of tumble,when using a wearable device to detect falling situation. In this paper,a PSO-SVM eigenvector fall detection algorithm based on human posture is proposed. Firstly,it collected the data of human body through the MEMS acceleration sensor node,and optimized the collected data by the conjugate gradient meth-ods to reduce the nonlinear error. Secondly,the support vector machine( SVM) is used to detect and classify the fall behavior,and the SVM parameters are optimized by Particle Swarm Optimization( PSO) algorithm to obtain the opti-mal classification model. According to analyze the collected data by SVM classification model,it can judge whether to fall;Finally,it can constructed the PSO-SVM eigenvector which fusing human posture angle to detect the specific information of fall process. The experimental results show that the proposed method attains a recognition rate of 95.5%,which can distinguish the other non-falls. The detection accuracy is higher than other methods, the root-mean-square error is smaller and the robustness is better.