测绘科学技术学报2025，Vol.41Issue(3)：282-288,7.DOI:10.3969/j.issn.1673-6338.2025.03.009
几何感知增强的RGB图像语义场景补全
RGB Image-based Semantic Scene Completion Using Geometric Perception Enhancement
摘要
Abstract
Semantic scene completion aims to improve the comprehensiveness and accuracy of environmental per-ception,which is of great significance for achieving safe autonomous driving.Although existing RGB image-based semantic scene completion methods have recognized the importance of restoring depth information,they have failed to fully explored the potential geometric information,which limits the improvement of the inference accuracy.To this end,a geometric perception module was proposed to be added to existing RGB image-based semantic scene completion methods,extracting geometric information from the pseudo point cloud obtained from depth estimation,and its output was embedded into hyperbolic space to enhance its geometric representation ability.The experimental results on the SemanticKITTI dataset prove the effectiveness of the proposed method.Compared with the baseline method,mIoU,the key evaluation metric,is improved by 2.5%,reaching the state-of-the-art performance of simi-lar methods.关键词
实时三维感知/语义场景补全/几何特征提取/双曲嵌入/自动驾驶/几何感知增强Key words
real-time 3D perception/semantic scene completion/geometric feature extraction/hyperbolic embed-ding/autonomous driving/geometric perception enhancement.分类
天文与地球科学引用本文复制引用
朱广阳,游雄,杨剑,李正瑜,魏宪..几何感知增强的RGB图像语义场景补全[J].测绘科学技术学报,2025,41(3):282-288,7.基金项目
国家自然科学基金重点项目(42130112) （42130112）
河南省中原学者游雄科学家工作室项目(2020478). （2020478）
