Deep Learning of Micro-Doppler Features for Aided and Unaided Gait Recognition

Abstract

Remote health monitoring is a topic that has gained increased interest as a way to improve the quality and reduce costs of health care, especially for the elderly. Falling is one of the leading causes for injury and death among the elderly, and gait recognition can be used to detect and monitor neuromuscular diseases as well as emergency events such as heart attack and seizures. In this work, the potential for radar to discriminate a large number of classes of human aided and unaided motion is demonstrated. Deep learning of micro-Doppler features is used with a 3-layer auto-encoder structure to achieve 89% correct classification, a 17% improvement in performance over the benchmark support vector machine classifier supplied with 127 pre-defined features.