Micro-Doppler Extraction of Pedestrian Limbs for High Resolution Automotive Radar

Abstract

Nowadays, micro-Doppler radar signatures are used for detection and classification of pedestrians. In order to further enhance pedestrian protection, pedestrian intention recognition plays a challenging role for which knowledge about the instantaneous movement of individual body parts is essential. This paper presents a new method for extracting body parts from radar measurements based on micro-Doppler, range and angle information. A signal processing chain for chirp sequence radar is described. After evaluating the range-Doppler frequency spectrum, the reflections belonging to the pedestrian are identified. A CLEAN algorithm is combined with a clustering algorithm to extract relevant scatters from different body parts. Finally, the azimuth angle for each relevant scatter is estimated. The proposed procedure is applied to real radar measurements of a walking pedestrian. The extracted scatters are compared to simultaneously measured references of several body parts from a motion capture system. The results show that the time-dependent motion behavior of the right and left leg, right and left arm and torso can be obtained from the radar data.