Abstract
A millimeter-wave radar is expected for autonomous driving and ADAS since it is less affected by weather and illumination conditions. However, it is difficult for the radar to precisely recognize shapes of the objects due to the low spatial resolution and noise. We dare to challenge reconstruction of the parking scene where they are required. Classes and shapes of various objects in the parking lot and street parking were estimated. We designed our original deep neural network and showed that it can be estimated with high accuracy. It was confirmed that ours is superior to the typical conventional networks.
