Disaggregation and detection of multiple automotive targets for millimeter-wave around-the-corner radar

Abstract

We experimentally demonstrate that around-the-corner radar (ACR) detection of automotive targets with millimeter-wave (77GHz) short-range radars can be leveraged for look ahead sensing into non-line-of-sight (NLOS) road turns for collision avoidance. However, there is significant overlap in the range-Doppler returns from multiple extended targets, in NLOS scenarios, due to multipath based ghost targets along the range and large micro-Doppler spreads from the vehicular wheels and arms and legs of pedestrians. As a result, it is challenging to resolve the returns from multiple targets before target detection and recognition. In this work, we propose the use of sparsity based dictionary learning techniques to separate and reconstruct the radar signatures of each target from the aggregated radar returns of multiple targets. Our experimental results, with a 77GHz automotive radar, with three target classes - car, pedestrian and bike - show a mean successful detection of 88.3% after disaggregation.