基于加速度时域特征和Ａｄａｂｏｏｓｔ.ＳＶＭ级联分类器的跌倒检测研究  

Fall Detection Based on Time Domain Features of Acceleration and Cascade-Adaboost-SVM Classifier

作　　者：田一明 王喜太[1,2] 杨鹏 耿艳利[1] Tian Yiming;Wang Xitai;Yang Peng;Geng Yan;Yang Peng(School of Control Science and Engineering,Hebei University of Technology,Tianjin 300130,China;Key Laboratory of Rehabilitation Aids Technology and System of the Ministry of Civil Affairs,Beijing Key Laboratory of Rehabilitation Technical Aids for Old.Age Disability,National Research Center for Rehabilitation Technical Aids,Beijing 100176,China)

机构地区：[1]河北工业大学控制科学与工程学院,天津300130 [2]国家康复辅具研究中心,北京市老年功能障碍康复辅助技术重点实验室,民政部康复辅具技术与系统重点实验室,北京100176

出　　处：《中南民族大学学报（自然科学版）》2018年第1期93-99,共7页Journal of South-Central University for Nationalities：Natural Science Edition

基　　金：国家科技支撑计划项目(2015BAI06B03);国家自然科学基金资助项目(61174009);河北省青年基金资助项目(F2016202327)

摘　　要：为了提高对跌倒检测的准确性和可靠性,提出了一种基于三轴加速度时域特征和Adaboost SVM级联分类器的跌倒检测方法.首先,利用滑动窗口法提取加速度信号的时域特征作为唯一特征向量,以提高系统检测的实时性;然后,对传统Adaboost算法的样本初始权值部分进行改进,使分类器学习到更多跌倒样本的信息,从而增强系统对跌倒的识别能力;最后,针对日常活动动作类(ADL)的数目远多于跌倒类而导致的数据集不平衡问题,构建了用于跌倒检测的Adaboost-SVM级联分类器,根据级联结构中每个Adaboost分类器所包含的弱分类器数量自动决定是否由SVM替换Adaboost分类器.利用UCI数据库中人体运动数据集进行了实验,结果表明:文中所提方法具有最高的跌倒检测率以及较为优秀的误报警率和准确率,并且证明了放置于胸部和腰部的加速度计能够对跌倒检测产生较好效果.In order to improve the accuracy and reliability in fall detection,this paper proposes a fall detection method based on time domain features extracted from triaxial accelerometer and cascade,AdaBoost,support vector machine(SVM)classifier.Firstly,sliding window method is used to extract the time domain features of acceleration signals as the only one feature vector.Secondly,on the basis of the improvement of the initial weight of the samples of the traditional Adaboost algorithm,a cascade,AdaBoost,SVM classifier is built to detect fall accidents for the dataset imbalance problem that the number of activities of daily life is far more than the fall,the algorithm can automatically determine whether to replace the AdaBoost classifier by support vector machine according to the number of weak classifier used in each Adaboost classifier to prevent overfitting of classifier.Lastly,the UCI database is used for the experiment.The results show that the proposed method have the highest detection rate as well as relatively perfect false alarm rate and accuracy rate.Furthermore,this paper proves that the triaxial accelerometers which are put on the chest and waist can produce optimal detection results.

关 键 词：跌倒检测 模式识别 集成学习 Adaboost级联算法 时域特征 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]