Design Of Fall Detection For The Elderly Based On The Wearable Sensor

Falls are one of the major causes of injury among the elderly.The wearable fall detection system can accurately detect the occurrence of a fall and alert caregivers or hospitals to offer rapid assistance,reducing fall-related injuries.However,most of the current studies on fall detection focus on the accuracy of detection,while ignoring the fact that battery life is also a key factor for the successful application of wearable fall detection system.This leads to the short battery life of the wearable device and the need for frequent charging or replacement of battery.Considering the power consumption and detection accuracy,this paper presents a wearable fall detection system,including a wearable device（TTXFD）,a master,and a fall detection algorithm combining threshold-based method（TBM stage）and machine learning technology（ML stage）.When an abnormal event occurs,the TTXFD exploits the TBM stage of the fall detection algorithm to preliminary identify the event.If the event is identified as a suspected fall event,a acceleration series is transmitted to the server.Finally,the master uses the ML stage of the fall detection algorithm to further detect the suspected fall to determine whether it is a fall.From the power consumption perspective,the TTXFD adopts various hardware-based methods to save power.At the same time,a feature selection approach with energy-saving effect is designed,that is,on the premise that the sensitivity of the TBM stage is 100%,the number of features at the TBM stage is minimized while the power consumption generated by the transmitted data is minimized.From the accuracy perspective,we use two machine learning techniques（convolutional neural network and support vector machine,CNN and SVM）at the ML phase to detect suspected fall events.The training results show that the ML stage based on SVM（accuracy=99.65%）outperforms the ML stage based on CNN（accuracy=99.60%）.Therefore,the master utilizes the ML stage based on SVM to improve the fall detection rate of the wearable fall detection system.The experimental results show that the proposed algorithm（TBM stage+ML stage based on SVM）has high sensitivity（98.56%）,specificity（97.76%）and accuracy（97.96%）on the testing set（offline testing）.In the real life（online testing）,the sensitivity,specificity and accuracy of the wearable fall detection system are 95.50%,93.60%and 94.44%,respectively.Moreover,a 500 mAh lithium battery provides about 62 days of power for the TTXFD.