정밀도로지도 갱신을 위한 YOLOv3 기반 도로시설물 객체 검출

Abstract

In recent decades, extensive technological efforts are able to improve autonomous driving. For example, High-Definition(HD) map has been constructed to ensure safety for autonomous driving. More specifically, a vehicle equipped with a Mobile Mapping System (MMS) collects data on road conditions first. Then humans check and modify any technical failure in order to create the final HD map. However, due to a lack of technological development, most jobs are done manually, which costs a tremendous amount of time and money. Therefore, a system, which automatically detects any changes on the road and corrects them synchronously on the map, is highly recommended. This study aims to test methodologies to develop an efficient change detection system in a road environment. To achieve this goal, this study first tried to detect objects in the video using the YOLOv3 algorithm, which is one of the object detection algorithms. Subsequently, via transfer learning, YOLOv3 which detect traffic sign, road sign objects and traffic light in HDmap was constructed. It achieves 56.5 in detecting the road facilities. By comparing the images before and after the road construction, It can detect any changes in road facilities and record road sections that needed corrections. A follow-up study would be conducted to compare the results of this study to results of studies employing state-of-the-art object detection algorithms or open set-based change detection.