Scholarlink

<p>A novel algorithm, called hierarchical transformer range image view network (HRNet), segmenting immobile elements for safe autonomous driving in urban with strong robustness and real-time performance is proposed in this paper. Among the algorithms for safe driving in urban autonomous driving, existing driving area detection tasks performed on image planes have observed image distortion and lack of depth information problems, and existing road boundary detection tasks performed in 3d space rely on rule-based methodologies, which often lack precision and struggle to achieve accurate classification in real-world scenarios. Therefore, this paper provides 3d accurate environmental perception information for identifying road boundaries and drivable areas with a dense 2D lidar range view. Our study proposes a hybrid architecture combining CNN and transformer based on Resnet-34 and ViT-B backbones for better processing global information, which differs from existing studies using CNN. Experimental results demonstrate our effectiveness on SemanticKITTI dataset.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

LiDAR-based semantic segmentation of immobile elements via hierarchical transformer for urban autonomous driving

  • 期刊论文

  • 发表日期：

    2025年5月31日

查看全文
  • Ayoung Lee,
  • Sungpyo Sagong,
  • Kyongsu Yi

摘要

A novel algorithm, called hierarchical transformer range image view network (HRNet), segmenting immobile elements for safe autonomous driving in urban with strong robustness and real-time performance is proposed in this paper. Among the algorithms for safe driving in urban autonomous driving, existing driving area detection tasks performed on image planes have observed image distortion and lack of depth information problems, and existing road boundary detection tasks performed in 3d space rely on rule-based methodologies, which often lack precision and struggle to achieve accurate classification in real-world scenarios. Therefore, this paper provides 3d accurate environmental perception information for identifying road boundaries and drivable areas with a dense 2D lidar range view. Our study proposes a hybrid architecture combining CNN and transformer based on Resnet-34 and ViT-B backbones for better processing global information, which differs from existing studies using CNN. Experimental results demonstrate our effectiveness on SemanticKITTI dataset.