Feature-based gesture classification by means of high resolution radar measurements

Continuously improving analog and digital hardware nowadays enables powerful waveforms at low costs and thus makes radar an attractive and cheap sensor for new fields of application. One such field is human machine interfaces (HMI). In this work, gesture classification based on high resolution radar measurements as new type of HMI is investigated. For this purpose, wideband radar measurements of human hand and body gestures at a center frequency of 60.5 GHz are recorded. As waveform, a sequence of linear frequency modulated (LFM) chirps with a bandwidth of 7 GHz is employed, allowing simultaneous high resolution measurements of range and radial velocity of multiple targets. Eight different gestures have been studied in this work. From the detector output, different features providing information of the relevant body parts are extracted using a proposed algorithm. These features of both training and test measurements are fed into the following classifiers: 1-nearest neighbor, p-nearest neighbor and polynomial classifier. It is shown, that the proposed radar signal processing, filtering and feature extraction methods yield very promising classification rates of over 95 % on the given data.