Human Activity and Posture Classification Using Single Non-Contact Radar Sensor

Radar has been proposed for monitoring the health of elderly patients in long term care because it is safe, non-contact and preserves the privacy of patients. Random body movements (RBM) obscure radar return signals making it difficult if not impossible to accurately estimate vitals. Activity classification is presented in this thesis as a pre-processing step for dealing with RBMs. Posture classification is presented in this thesis for assistance in preventing falls. Two popular radar architectures- continuous wave (CW) Doppler and ultra-wideband (UWB) are investigated in this thesis. Activity classification is performed with 92% average accuracy with CW and 86% with UWB. Posture Classification is performed with 64% average accuracy with CW and 85% with UWB. An occupancy detection algorithm was also developed for UWB and achieved 88% average accuracy. The contribution of this thesis is a proposed hierarchical processing approach for both radar types capable of dealing with moving subjects.