Scholarlink

<p>Object SLAM plays a crucial role in enhancing the environmental perception capabilities of SLAM systems. However, most current research on object SLAM remains confined to small indoor scenes, restricting its applicability in broader fields such as autonomous robotics and autonomous driving.</p><p> These fields demand robust SLAM solutions capable of handling complex outdoor environments. To overcome the challenges posed by outdoor scenes, such as small viewpoint changes and insufficient object observation, we have enhanced the quadric initialization and association module within object SLAM. By utilizing outdoor depth data collected from a full-scene depth camera, our approach enables more robust object reconstruction in outdoor settings. In addition, we refined the relocalization module by leveraging the high-quality object maps produced by our system. This enhancement considerably improves the long-term success rate of relocalization in dynamic outdoor environments. Experimental results validate that our system outperforms existing object SLAM methods, achieving at least a 10% improvement in outdoor object mapping accuracy and over a 5% enhancement in relocalization, demonstrating its robustness and effectiveness in challenging outdoor environments.</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

Enhancing object SLAM for outdoor environments: robust reconstruction and relocalization

  • 期刊论文

  • 发表日期：

    2026年5月20日

查看全文
  • Xuan An,
  • Kang Li,
  • Li’an Wang,
  • Yujie Ji,
  • Yuxuan Wu,
  • Zhaoyuan Ma

摘要

Object SLAM plays a crucial role in enhancing the environmental perception capabilities of SLAM systems. However, most current research on object SLAM remains confined to small indoor scenes, restricting its applicability in broader fields such as autonomous robotics and autonomous driving.

These fields demand robust SLAM solutions capable of handling complex outdoor environments. To overcome the challenges posed by outdoor scenes, such as small viewpoint changes and insufficient object observation, we have enhanced the quadric initialization and association module within object SLAM. By utilizing outdoor depth data collected from a full-scene depth camera, our approach enables more robust object reconstruction in outdoor settings. In addition, we refined the relocalization module by leveraging the high-quality object maps produced by our system. This enhancement considerably improves the long-term success rate of relocalization in dynamic outdoor environments. Experimental results validate that our system outperforms existing object SLAM methods, achieving at least a 10% improvement in outdoor object mapping accuracy and over a 5% enhancement in relocalization, demonstrating its robustness and effectiveness in challenging outdoor environments.