Indoor Tracking And Evaluation Of Activity Of Daily Living For Elderly Based On UWB Radar

While the aging of the population has brought challenges to countries around the world,it has also promoted the emergence of the concept of ambient assisted living and the development of smart elderly care.The establishment of an indoor daily living monitoring system,which is based on environment and biological intelligence,could feed back the physical condition of the elderly to the remote family members in time,so as to realize auxiliary care.In order to avoid privacy exposure and physical intrusion caused by traditional sensing methods,such as cameras and wearable devices,ultra-wideband(UWB)radar may be a viable alternative.In this research,UWB radar is used for indoor human target detection and tracking,motion recognition,motion parameter extraction,and preliminary evaluation of activity of daily living.The main works are as follows:1.As for the multi-point scattering and range expansion characteristics of human target,a target detection method based on baseband signals is proposed,which effectively reduces the number of measurements,data storage space and algorithm processing time,and further improves real-time performance;2.According to target detection results,a measurement clotting and jumping-window starting method is proposed,which is supposed to determine the number of targets and their initial states,so as to achieve single target and multi-target tracking respectively.The trajectory update time is0.1 s,while the RMSE of tracking is less than 0.3 m;3.To overcome the limitation in application of existing activity of daily living evaluation tools,the SWS tool is established,which uses the completion time of a single sitting,3 m walking,and a single squat.Its specific process includes two stages respectively,which are named motion recognition and motion parameter extraction;4.In the motion recognition stage,according to Doppler effect result from human body's action,the radar signals are converted into time-range-Doppler spectrogram.Then,Convolutional auto-encoder is applied to feature extraction,and QRNN is used for timing modeling.This model has an accuracy of 96.58% for the recognition of 8 common indoor motions,and also has advantages in inference time;5.In the motion parameter extraction stage,for squatting and sitting,the radar signals are converted into time-Doppler spectrogram,and the motion's duration is extracted by velocity constraints.As for walking,time-Doppler spectrogram and target tracking trajectory are combined for comprehensive judgment,to output the required time for actual displacement of 3 m.In the end,SWS scoring standard is applied to reflect ability of activity of daily living.