Scholarlink

With the advancement and development of technology, unmanned platforms have shown wide prospects in fire search and rescue. To solve the problems of high dynamic target loss and repositioning failure in indoor fire smoke interference scenarios, this paper proposes a simultaneous localization and mapping (SLAM) method applied to unmanned SAR platforms. In order to improve the clarity of the unmanned platform sensor imaging picture, the method is combined with a Deep Multi-Patch Hierarchical Network (DMPHN) to repair fogged images for subsequent precise localization. In this paper, a SAR-SLAM algorithm based on YOLOv8s is designed to enhance the precision of localization estimation and mapping of unmanned platforms in indoor dynamic environments. This algorithm is able to use to segment dynamic image instances using YOLOv8s. Next, the dynamic feature points are deleted and the image is fed into the tracking thread of ORB-SLAM3 to better realize the path planning and navigation of the unmanned search and rescue platform. The experimental results show that the performance of this SAR-SLAM under the publicly available dataset TUM significantly enhances the precision of the localization estimation compared to ORB-SLAM3, with an average improvement rate of 91.83%.

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

Dynamic SLAM Algorithm for Semantic-Driven Unmanned Platforms in Multi-factor Environments

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Jingyi Wei,
  • Zhenpeng Yin,
  • Jiajun Lin,
  • Hongyang Bai,
  • Yuman Yuan,
  • Xiaochuan Zhao,
  • Jinzhu Shi

摘要

With the advancement and development of technology, unmanned platforms have shown wide prospects in fire search and rescue. To solve the problems of high dynamic target loss and repositioning failure in indoor fire smoke interference scenarios, this paper proposes a simultaneous localization and mapping (SLAM) method applied to unmanned SAR platforms. In order to improve the clarity of the unmanned platform sensor imaging picture, the method is combined with a Deep Multi-Patch Hierarchical Network (DMPHN) to repair fogged images for subsequent precise localization. In this paper, a SAR-SLAM algorithm based on YOLOv8s is designed to enhance the precision of localization estimation and mapping of unmanned platforms in indoor dynamic environments. This algorithm is able to use to segment dynamic image instances using YOLOv8s. Next, the dynamic feature points are deleted and the image is fed into the tracking thread of ORB-SLAM3 to better realize the path planning and navigation of the unmanned search and rescue platform. The experimental results show that the performance of this SAR-SLAM under the publicly available dataset TUM significantly enhances the precision of the localization estimation compared to ORB-SLAM3, with an average improvement rate of 91.83%.