以馬路為跑道之定翼無人機跑道自動辨識系統

降落是無人機飛行中最具挑戰性的階段。缺乏高效的系統導致降落事故屢見不鮮，造成機上硬件損壞。為了解決這個問題，本研究採用YOLO(You Only Look Once) 深度學習網路以及影像處理技術，將馬路視為定翼無人機的降落跑道。首先，我們搜集了相關的圖像並進行標註，以建立訓練數據庫，確保圖像涵蓋各種視角下的馬路場景，從而使模型能夠適應多樣的環境。接著，我們運用YOLOv4 物件辨識演算法進行模型訓練，同時測試了模型的預測結果。再將模型的預測結果透過影像處理技術，像是Canny邊緣檢測和Hough轉換找尋降落跑道的中線。最終目標是開發一個有效的跑道自動辨識系統，有助於無人機在著陸時更準確地識別合適的著陸場地，將馬路視為其中一個可行的降落區域。

關鍵字

YOLOv4 ； Canny 邊緣檢測； Hough 轉換

並列摘要

Landing is the most challenging phase of drone flight. The lack of efficient systems frequently leads to landing accidents, causing hardware damage. To address this issue, this study employs the YOLO (You Only Look Once) deep learning network and image processing techniques, considering roads as runways for fixed-wing drones. First, we collected relevant images and annotated them to establish a training database, ensuring the images cover road scenes from various perspectives so the model can adapt to diverse environments. Then, we trained the model using the YOLOv4 object detection algorithm and tested the model's predictions. The model's prediction results were further processed using image processing techniques, such as Canny edge detection and Hough transform, to find the centerline of the landing runway. The ultimate goal is to develop an effective runway automatic detection system that aids drones in accurately identifying suitable landing sites, with roads being one of the feasible landing areas.