论文部分内容阅读
Conventional activity recognition can be effectively captured by accelerometer data,especially for the fall caused injuries that are seriously great threats to the elderly people.Accordingly,timely detecting and monitoring the occurrence of fall accidents are extremely meaningful to help the injured person receive the first aid.However,most existing approaches for building classifiers struggle to balance high accuracy and low falsealarm rate in large-scale wearable computing applications.To address this,we propose FallSense,a novel fall detection system using eyeglass-watch wearables over a multi-sensor fusion based two-stage recognition method.To the best of our knowledge,this is the first work to both eyeglass and watch wearables based fall detecting and monitoring.Furthermore,our system present a completely unobtrusive monitoring way to deploy and equip acceleration sensors on the eyeglass and watch,where the elderly people often wear presbyopic glasses and watches in the daily life.Experimental results validate the effectiveness of the proposed method and show that ELM classifier is effective to implement on resource-constrained wearable devices,and adding SVM classifier for result refining we can finally obtain higher detection accuracy and lower false-alarm rate simultaneously than the state-of-the-art fall detection approaches,which greatly helps the healthcare and monitoring especially for the elderly people in terms of fall detection,fall prevention,first-aid etc.