Scholarlink

A global navigation Global Navigation Satellite System (GNSS)-free navigation system based on the integration of boundary regularization, Progressive Hough Transform, and YolactEdge to improve the results of runway detection and localization has been proposed to keep the performance of navigation system, which is affordable for onboard embedded systems to perform real-time applications of unmanned aerial vehicles (UAVs). A monocular camera is used as the sensor, and the embedded system is employed for object detection, image segmentation, and image post-processing to determine the relative distance between the runway and the aircraft. The one-stage image deep learning segmentation algorithm, YolactEdge, is utilized to extract the pixel coordinates of the runway contours in the images. Plucker method, an image localization algorithm, is then applied to calculate the relative relationships between the runway and owned aircraft. Therefore, the Douglas-Peucker method is employed to find closely fitted line segments. Finally, the Progressive Hough Transform is applied to perform edge detection on the binary image. In this research, the feasibility of the proposed methods is validated through flight simulations in the X-Plane flight simulator. Additionally, a real flight experiment was conducted for image recognition and image-based positioning.

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

Development of a Vision-Based Navigation and Landing System Based on Deep Learning and Image Segmentation Methods with Runway Information for the Landing of a Fixed-Wing UAV

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Wen-Hao Liao,
  • Ying-Chih Lai

摘要

A global navigation Global Navigation Satellite System (GNSS)-free navigation system based on the integration of boundary regularization, Progressive Hough Transform, and YolactEdge to improve the results of runway detection and localization has been proposed to keep the performance of navigation system, which is affordable for onboard embedded systems to perform real-time applications of unmanned aerial vehicles (UAVs). A monocular camera is used as the sensor, and the embedded system is employed for object detection, image segmentation, and image post-processing to determine the relative distance between the runway and the aircraft. The one-stage image deep learning segmentation algorithm, YolactEdge, is utilized to extract the pixel coordinates of the runway contours in the images. Plucker method, an image localization algorithm, is then applied to calculate the relative relationships between the runway and owned aircraft. Therefore, the Douglas-Peucker method is employed to find closely fitted line segments. Finally, the Progressive Hough Transform is applied to perform edge detection on the binary image. In this research, the feasibility of the proposed methods is validated through flight simulations in the X-Plane flight simulator. Additionally, a real flight experiment was conducted for image recognition and image-based positioning.