Scholarlink

<p>This paper presents Ellipsoid-SLAM, an object-level Simultaneous Localization and Mapping (SLAM) method specifically designed to enhance the accuracy and robustness of camera pose measurement in dynamic scenes. To address the challenges posed by dynamic objects, we introduce a novel approach that represents detected instance objects as 3D ellipsoids and constructs an object map. By leveraging motion detection, we identify dynamic objects and measure their trajectories within the 3D map. Additionally, we propose an enhanced relocalization method that jointly optimizes static objects and internal 3D points to accurately measure the camera pose after tracking loss. Experimental results on the TUM and BONN datasets demonstrate that Ellipsoid-SLAM achieves higher camera pose measurement accuracy compared to state-of-the-art methods, particularly in dynamic scenes. Our approach not only tracks dynamic objects but also eliminates unstable feature points on these objects, resulting in more precise camera pose estimation. The source code and datasets used in this study are available at <a href="https://github.com/CodingMaplee/DEllipsoid-SLAM">,</a> complete with comprehensive usage guidelines to facilitate replication and evaluation of our results.</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

Ellipsoid-SLAM: enhancing dynamic scene understanding through ellipsoidal object representation and trajectory tracking

  • 期刊论文

  • 发表日期：

    2025年2月5日

查看全文
  • Haowei Zhu,
  • Suqin Bai,
  • Jinlong Shi,
  • Jiawen Lu,
  • Xin Zuo,
  • Shucheng Huang,
  • Xu Yao

摘要

This paper presents Ellipsoid-SLAM, an object-level Simultaneous Localization and Mapping (SLAM) method specifically designed to enhance the accuracy and robustness of camera pose measurement in dynamic scenes. To address the challenges posed by dynamic objects, we introduce a novel approach that represents detected instance objects as 3D ellipsoids and constructs an object map. By leveraging motion detection, we identify dynamic objects and measure their trajectories within the 3D map. Additionally, we propose an enhanced relocalization method that jointly optimizes static objects and internal 3D points to accurately measure the camera pose after tracking loss. Experimental results on the TUM and BONN datasets demonstrate that Ellipsoid-SLAM achieves higher camera pose measurement accuracy compared to state-of-the-art methods, particularly in dynamic scenes. Our approach not only tracks dynamic objects but also eliminates unstable feature points on these objects, resulting in more precise camera pose estimation. The source code and datasets used in this study are available at , complete with comprehensive usage guidelines to facilitate replication and evaluation of our results.