Micro-Doppler Gesture Recognition using Doppler, Time and Range Based Features

This paper presents micro-Doppler analysis and classification results from radar measurements of various hand gestures. A new database of 6 individuals completing 4 separate gestures with over 3,000 repetitions was recorded using a 24 GHz Ancortek radar system. The micro-Doppler signatures from these gestures were generated, features extracted and multiple different classifiers applied to this gesture data. A typical micro-Doppler classification process aims to use either a single range bin of data, average over a series of range bins or align all the target signal to a single bin. Different to previous techniques, the paper presents a method that uses multiple ranges bins to produce a spectrogram per range bin in order to represent the observed gesture over all four dimensions of time, Doppler, space and polarization. A comparison of the traditional and the newly proposed technique is shown and the improvements demonstrated are observed to be significant.