Scholarlink

This paper proposes a robust VSLAM system for environments with dynamic objects. To reduce the impact of dynamic elements in VSLAM, we integrate a deep learning-based instance segmentation method into visual odometry. Afterward, image processing is applied to the resulting mask image. We validate this method by using the TUM dataset, and the result shows that the method greatly reduces tracking errors and drift, improving the accuracy when dealing with dynamic object edges.

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

RGB-D SLAM in Indoor Dynamic Environment Based on Instance Segmentation

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Jingxuan Xiang,
  • Yujuan Wang,
  • Ruping Ceng,
  • Qing Chen,
  • Ziguo Liu,
  • Zheng Zhou,
  • Qing Zhang,
  • Yufan Wang

摘要

This paper proposes a robust VSLAM system for environments with dynamic objects. To reduce the impact of dynamic elements in VSLAM, we integrate a deep learning-based instance segmentation method into visual odometry. Afterward, image processing is applied to the resulting mask image. We validate this method by using the TUM dataset, and the result shows that the method greatly reduces tracking errors and drift, improving the accuracy when dealing with dynamic object edges.