Human Motion Identity using Machine Learning on Spectral Analysis of RSS Signals

Human motion identity based on user interaction with wireless signals in an indoor environment can be of great assistance supporting health and welfare services in society. For example, a system can detect an abnormal state or behavior as well as monitor the security of the elderly and infirm. This paper describes a methodology for motion identity based on the passive interaction of wireless LAN signals. The received signal strength (RSS) is recorded for several movement directions as well as objects placed relative to the transmitter and receiver. The technique uses the continuous wavelet transform (CWT) to generate a scalogram time-frequency intensity image from the RSS data and a convolutional neural network (CNN) is then trained to recognize the unique spectral features of each image and enable the motion classification.