传感技术学报2025，Vol.38Issue(10)：1793-1799,7.DOI:10.3969/j.issn.1004-1699.2025.10.009
基于结构重参数化的老人跌倒检测算法研究
Research on Fall Detection Algorithm for the Elderly Based on Structural Reparameterization
摘要
Abstract
Developing a reliable wearable fall detection device is essential for mitigating injuries among the elderly resulting from fall.Most of current deep learning models for elderly fall detection struggle to balance accuracy with computing cost effectively.A novel fall detection classification model utilizing structural reparameterization is studied and designed to solve the above problems.Initially,the moment of a is identified fall based on peak combined acceleration,which is used as the central point for data windowing.Two seconds data around this point are then captured as the model's input data.A multi-branch convolutional kernel,serving as the backbone for feature extraction,is developed.To enhance efficiency,the Ghost Module is integrated,streamlining the model and minimizing computational demands.Structural reparameterization differentiates the model's training from its testing,allowing for the extraction of acceleration,gyroscope,and Euler angle features via the multi-branch convolutional kernel during training.During testing,this multi-branch structure condenses into a single-branch framework,reducing computational load and improving inference speed while maintaining model performance.Results of ex-perimental validation on both KFall and SisFall datasets demonstrate the model's excellence,achieving accuracy of 99.60%and 98.67%,specificity of 99.56%and 98.58%,sensitivity of 99.64%and 98.72%,respectively.关键词
跌倒检测/轻量化模型/结构重参数化/可穿戴设备/惯性传感数据Key words
fall detection/lightweight model/structural reparameterization/wearable device/inertial data分类
计算机与自动化引用本文复制引用
宋炜,周辰,潘巨龙..基于结构重参数化的老人跌倒检测算法研究[J].传感技术学报,2025,38(10):1793-1799,7.基金项目
浙江省基础公益研究计划项目(LGF21F020017) （LGF21F020017）
