Scholarlink

<p>In the field of intelligent mobile robots, simultaneous localization and mapping (SLAM) technology is crucial for navigation in complex dynamic environments. However, traditional visual SLAM frameworks, while performing excellently in static environments, often face challenges such as pose estimation drift and tracking loss in dynamic environments, primarily due to interference from moving objects. To overcome these obstacles, this paper proposes a visual SLAM algorithm based on ORB-SLAM3. The algorithm first introduces an improved line feature extraction technique to enhance the robustness of line features in low-texture environments. Additionally, the improved YOLOv8 is utilized to detect and eliminate dynamic points in real-time, thereby reducing the impact of dynamic objects on pose estimation accuracy. The improved YOLOv8 model reduces computational complexity by integrating the GhostConv module and enhances feature representation by introducing the Global Attention Mechanism (GAM). Furthermore, depth data are fused to construct a dense point cloud map. Experiments were conducted on the TUM dataset, the EuRoC dataset, and real-world environments, including localization accuracy and real-time performance analysis. The results indicate that, compared to ORB-SLAM3, the proposed algorithm achieves significant improvements in overall performance. These advancements greatly enhance the navigation precision and reliability of robots in dynamic 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

Visual SLAM Based on Line Feature Extraction and Dynamic Object Removal in Dynamic Environments

  • 期刊论文

  • 发表日期：

    2026年1月5日

查看全文
  • Ruizhen Gao,
  • Qishuo Han,
  • Guoguang Li,
  • Ziyue Zhao,
  • Beibei Wang,
  • Guocheng Zhang

摘要

In the field of intelligent mobile robots, simultaneous localization and mapping (SLAM) technology is crucial for navigation in complex dynamic environments. However, traditional visual SLAM frameworks, while performing excellently in static environments, often face challenges such as pose estimation drift and tracking loss in dynamic environments, primarily due to interference from moving objects. To overcome these obstacles, this paper proposes a visual SLAM algorithm based on ORB-SLAM3. The algorithm first introduces an improved line feature extraction technique to enhance the robustness of line features in low-texture environments. Additionally, the improved YOLOv8 is utilized to detect and eliminate dynamic points in real-time, thereby reducing the impact of dynamic objects on pose estimation accuracy. The improved YOLOv8 model reduces computational complexity by integrating the GhostConv module and enhances feature representation by introducing the Global Attention Mechanism (GAM). Furthermore, depth data are fused to construct a dense point cloud map. Experiments were conducted on the TUM dataset, the EuRoC dataset, and real-world environments, including localization accuracy and real-time performance analysis. The results indicate that, compared to ORB-SLAM3, the proposed algorithm achieves significant improvements in overall performance. These advancements greatly enhance the navigation precision and reliability of robots in dynamic environments.