摘要
目的:为了提高跌倒分类任务的精度,同时保证跌倒检测的实时性。方法:提出了一种融合Lasso回归和轻量级梯度提升机(Lightweight Gradient Boosting Machine,LightGBM)的跌倒检测算法Lasso-LGB。该方法首先利用Lasso回归算法选取跌倒数据特征向量中的主要特征;然后用LightGBM算法来检测判别人体的跌倒行为或日常生活行为。结果:通过两个公开的跌倒数据集进行算法验证,表明Lasso-LGB跌倒算法能更准确地检测跌倒行为,使跌倒检测的误报率和漏报率大大降低。结论:提出的Lasso-LGB算法实现了高精度的跌倒行为检测以及准实时的运行时间需求。
Aims:This paper aims to improve the accuracy of fall classification and ensure the real-time performance of fall detection.Methods:A fall detection algorithm(Lasso-LGB)combining Lasso regression and the Lightweight Gradient Boosting Machine(LightGBM)was proposed.Lasso regression was used to select the main features in the feature vectors of fall data.The LightGBM algorithm was used to judge whether the human behavior was fall behavior or daily living activity.Results:The algorithm was verified through two public fall datasets.The results showed that the Lasso-LGB algorithm could detect fall behavior accurately;and the false alarm rate and missing alarm rate were reduced.Conclusions:The proposed fall detection algorithm achieves the requirements of high-precision detection of fall behavior and real-time detection.
作者
段美玲
潘巨龙
DUAN Meiling;PAN Julong(College of Information Engineering,China Jiliang University,Hangzhou 310018,China)
出处
《中国计量大学学报》
2021年第1期67-73,117,共8页
Journal of China University of Metrology
基金
浙江省基础公益研究计划项目(No.LGF21F020017)。
关键词
跌倒检测
Lasso回归
机器学习
时域特征
轻量级梯度提升机
fall detection
Lasso regression
machine learning
time-domain features
lightweight gradient boosting machine