Marker-based Localization for Automated Parking Using Automotive Radar Point Cloud

Automated parking requires a high localization precision since the vehicles have to move in narrow spaces [1]. Recently, 77 GHz automotive radar has been widely applied to autonomous driving, due to its robustness to different lighting and weather conditions. We propose a marker-based localization method for automated parking using a 77 GHz automotive radar. The proposed method robustly detects the markers from the sparse radar point clouds by building local maps and accurately estimates the ego-pose with a learning-based registration method. The real-world evaluations show that our method outperforms the traditional methods and can achieve a centimeter-level positioning accuracy in various parking scenarios.