基于SVM与多数据集的摔倒检测方法研究

Research on Fall Detection Method Based on SVM and Multiple Data Sets

摔倒检测系统能够实时监测用户的运动状态,当用户摔倒时,自动发出警报,让用户得到及时的救治。对于摔倒检测方法的研究通常采用自建数据集进行验证实验,然而,数据集的差异影响实验结果的好坏,实验结果无法进行比较,影响实验结果的可信度。文章使用三个基于加速度数据的摔倒检测公开数据集,采用基于运动方向特征及支持向量机(SVM)的方法进行分类实验。分类算法在三个数据集上的实验结果为:Mobi Act数据集的灵敏度与特异性为97.41%和98.10%;t Fall数据集的灵敏度与特异性为96.55%和97.86%;Sis Fall数据集的灵敏度与特异性为98.77%和99.03%。

Fall detection system can monitor the user＇s real-time movement.When the user Ms,it can automatically send an alarm so that user receives timely treatmentResearch on fall detection methods usually uses self-built data sets for verification experimentsesearchers.The dierences between the data set aect the experimental results.The experimental results of dierent articles can not be compared.Di？erent datasets in？uence the credibility of the ejqjerimental results.In this paper,three public datasets based on acceleration data are used.The feature of motion direction and support vector machine（SVM）are used to do classi cation experiments.Our experimental results on three datasets：The sensitivity and sped？city of the MobiAct dataset were 97.41% and 98.10%.The sensitivity and sped city of the tFall dataset werc 96.55% and 97.86%.The sensitivity and sped city of tiie SisFall dataset were 98.77% and 99.03%.