도심 환경 속 판단 보조를 위한 교통섬 지형분할 및 객체분류

Abstract

In the urban environment, terrains such as a “traffic island” are present for the convenience and safety of pedestrians. However, such terrains cause decision errors in the autonomous system. In the existing autonomous vehicles, a method of stopping or decelerating the vehicle for safety when an object is detected on the driving path or around the vehicle is applied. However, such methods not only make it difficult for the autonomous vehicles to drive smoothly in unusual situations such as a traffic island but also cause traffic jam with the possibility to cause traffic accidents in the urban environment. To mitigate this problem, we segmented the traffic island and roadway through semantic segmentation. Pedestrians on the traffic island are classified as the “safety group,” and the other areas are termed as the “non-safety group,” providing an efficient method for stable autonomous driving decisions in the Pangyo area.