Universal Image Segmentation Framework on Highresolution Automotive Radar Map

A universal image segmentation framework, which can be applied to various high-resolution automotive radar imagery produced by different beamforming strategies, is expected in the radar community to provide robust support to the development of autonomous driving. This paper estimates the universality of the segmentation framework, which is developed based on radar data produced by the mechanical steer beamforming, by directly implementing it onto another high-resolution radar imagery produced by the beamforming strategy of MIMO Doppler beam sharpening (DBS). The comparison of the distribution features of two parts of data shows that the return power level shift caused by the resolution difference is the major factor that needs to be compensated for the framework transfer implementation. The details of the universal segmentation framework are given to show that this can significantly simplify the complicated manual labelling and feature extraction. The segmentation results are discussed with the analysis of the performance and the potential future work.