Shadow-Robust Semantic Segmentation for Autonomous Navigation in Walking Space
Kota Hayashi, Hiroaki Nakamichi, Hiroshi Yoshitake, Motoki Shino · 2024
Extensive research has been conducted on autonomous navigation systems for last-mile outdoor transportation. A method in previous work utilizes semantic segmentation to classify driving environments into classes such as recommended traversable areas for autonomous navigation following traffic rules in walking space. However, the presence of shadows on the road surface negatively impacts the accuracy of semantic segmentation, consequently leading to failures in autonomous navigation. This paper proposes an approach to demonstrate shadow-robust semantic segmentation, employing a multi-head semantic segmentation model. By utilizing our proposed method, autonomous navigation following traffic rules in walking space was successful under conditions with shadows.