인문학
사회과학
자연과학
공학
의약학
농수해양학
예술체육학
복합학
DBpia AI
AI 에이전트
AI 아이디어
AI 리더
AI 뷰어
Citeasy
크롬 서지관리 다운로드
지원사업
학술연구/단체지원/교육 등 연구자 활동을 지속하도록 DBpia가 지원하고 있어요.
커뮤니티
연구자들이 자신의 연구와 전문성을 널리 알리고, 새로운 협력의 기회를 만들 수 있는 네트워킹 공간이에요.
피인용 0
초록·키워드
Semantic segmentation of LiDAR point clouds plays a crucial role in autonomous driving by providing fine-grained scene understanding for safe decision-making. However, obtaining dense 3D annotations is prohibitively expensive due to the significant manual labeling effort required. To mitigate this challenge, a practical alternative is to project 2D segmentation results from cameras onto 3D point clouds, thereby generating pseudo labels. While promising, this approach faces two major limitations: (i) the mismatch in resolution and field-of-view (FOV) between LiDAR and cameras, which leaves many 3D points unlabeled, and (ii) the presence of occlusions when aggregating multi-frame LiDAR data into a reference frame, leading to erroneous label propagation. In this work, we propose an occlusion-aware pseudo ground truth generation framework that employs Z-buffering to handle depth conflicts during 2D-to-3D unprojection. Our method first aligns LiDAR point clouds across multiple frames using ego-poses, then applies a depth-priority mechanism to discard occluded points before transferring camera-based labels. This design produces more reliable pseudo labels while maintaining low annotation costs. We evaluate our framework on the Waymo Open Dataset, showing that the generated pseudo ground truth aligns closely with real-world urban driving scenes. Experiments demonstrate that our approach not only reduces labeling effort but also improves the quality of supervision for training robust LiDAR semantic segmentation models.
본문·목차
인공지능 문자 인식 모델을 통해 추출된 텍스트로, 일부 오타나 오류가 포함될 수 있으나 지속적으로 개선 중입니다.
오류를 발견하셨다면 해당 부분을 드래그한 후 ' 를 통해 신고해주세요.
오류를 발견하셨다면 해당 부분을 드래그한 후 ' 를 통해 신고해주세요.
최근 본 자료 전체보기
UCI(KEPA) : I410-151-26-02-095655328
