Scholarlink

<p>Accurate distinction between static and moving objects is crucial for precise localization and reliable environmental mapping in autonomous driving. This paper presents a novel dual-encoder, single-decoder network architecture that separately processes static and moving features. In the static encoder, we develop a feature extraction technique to detect more discriminative geometric details, while an attention mechanism in the dynamic encoder is employed to extract and track moving features within temporal variations. Additionally, we implement a weight-sharing strategy to effectively fuse feature weights across branches. Extensive experiments on the SemanticKITTI dataset demonstrate that our method outperforms existing techniques in moving object detection with a 73.7% Intersection over Union of moving objects (mIoU) by reducing false positives.</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

An efficient network for 3D point cloud moving object detection in autonomous driving environments

  • 期刊论文

  • 发表日期：

    2025年9月22日

查看全文
  • Jiaping Chen,
  • Kebin Jia,
  • Yuxuan Zhao

摘要

Accurate distinction between static and moving objects is crucial for precise localization and reliable environmental mapping in autonomous driving. This paper presents a novel dual-encoder, single-decoder network architecture that separately processes static and moving features. In the static encoder, we develop a feature extraction technique to detect more discriminative geometric details, while an attention mechanism in the dynamic encoder is employed to extract and track moving features within temporal variations. Additionally, we implement a weight-sharing strategy to effectively fuse feature weights across branches. Extensive experiments on the SemanticKITTI dataset demonstrate that our method outperforms existing techniques in moving object detection with a 73.7% Intersection over Union of moving objects (mIoU) by reducing false positives.