四旋翼無人機結合視覺辨識障礙物與自主路徑導航

Abstract

本研究實現一套具備即時避障與自主導航能力之四旋翼無人機系統，整合視覺辨識、空間分析與路徑規劃技術，以提升無人機在複雜環境中的適應性。首先，在障礙物辨識方面，應用了Farnebäck 光流法，並結合光流補償提升移動物體偵測準確度。隨後，採用 DBSCAN 類聚法與區塊劃分方式，分析影像中障礙區域的分佈與面積，進一步轉換為 3D 空間中的實體位置資訊。在路徑規劃部分，使用 BFS 廣度優先搜尋演算法進行障礙區域探索，並結合反投影法將 2D 影像座標轉換至 3D 空間中，提供動態避障資訊。針對整體導航策略，採用快速探索隨機樹演算法（RRT*）產生初始路徑，再依序跟隨生成的航點進行飛行引導。在導航過程中，利用機載鏡頭即時取得影像，建立光流場以進行障礙物偵測，並即時分析四旋翼飛行器與障礙物之間的相對關係，實現動態調整與安全避障。實驗結果驗證，本研究所提出之系統可穩定進行視覺辨識、動態避障與空間導航，具有良好之實用性。This study presents the implementation of a quadrotor unmanned aerial vehicle (UAV) system capable of real-time obstacle avoidance and autonomous navigation. The system integrates visual perception, spatial analysis, and path planning techniques to enhance UAV adaptability in complex environments.For obstacle detection, the Farnebäck optical flow method with flow compensation is applied to enhance moving object detection. DBSCAN clustering and grid-based segmentation are used to analyze obstacle areas, which are then mapped into 3D space. A breadth-first search (BFS) algorithm explores obstacle regions, and a back-projection method converts 2D image coordinates into 3D positions for dynamic obstacle avoidance. The Rapidly-exploring Random Tree Star (RRT*) algorithm generates the initial navigation path. During the navigation process, real-time images are captured by the onboard camera to construct an optical flow field for obstacle detection. The relative relationship between the quadrotor UAV and obstacles is analyzed in real time, enabling dynamic adjustment and safe obstacle avoidance.Experimental results show that the system achieves stable visual recognition, obstacle avoidance, and navigation performance, demonstrating strong potential for practical UAV applications.