無人機的自主飛行避障技術中，常見的解決方案包括紅外線、超音波、雷射 和視覺避障技術。本論文選擇了單目視覺技術，並運用了 DIS（Dense Inverse Search）光流演算法來計算光流影像，結合IMU數據，通過Flowdep演算法將光 流影像轉換為精確的深度數據。本研究不僅驗證了該方法在理想場景下的效果， 如帶有紋理背景的場景，還探討了該方法在不利環境（如純戶外、有天空及地平 線的環境）下的應用潛力。為了提高深度估算的穩定性，對戶外環境中的光流影 像應用了濾波器，以過濾雜訊並減少錯誤估算。 本研究還提出了一種基於畫面分割的避障策略，即只針對選定區域進行障礙 物檢測，避免了整個畫面帶來的運算負擔及錯誤判斷風險。這種分區判斷方法大 大提高了運算效率，並保持了避障系統的準確性和可靠性。經過多次實驗測試， 該避障策略在低速飛行情境下表現出穩定且高效的避障能力，尤其是在樹林等複 雜環境中的應用效果尤為顯著。這些結果證實了基於Flowdep的視覺避障方案的 可行性，並為未來的無人機自主避障技術提供了有價值的參考，尤其是在戶外複 雜環境中的應用潛力。

In autonomous flight obstacle avoidance technology for drones, common solutions include infrared, ultrasonic, laser, and vision-based methods. This paper selects monocular vision technology and utilizes the Dense Inverse Search (DIS) optical flow algorithm to compute optical flow images. Combined with IMU data, the optical flow images are converted into precise depth information using the Flowdep algorithm. This study not only validates the effectiveness of this method in ideal scenarios, such as textured backgrounds, but also explores its potential applications in challenging environments, such as outdoor scenes with sky and horizon. To enhance the stability of depth estimation, a filter is applied to the optical flow images in outdoor environments to reduce noise and minimize erroneous estimations. Additionally, this research proposes an obstacle avoidance strategy based on image segmentation, where only selected areas are detected for obstacles, avoiding the computational burden and misjudgment risks associated with processing the entire frame. This region-based detection method significantly improves computational efficiency while maintaining the accuracy and reliability of the obstacle avoidance system. Through multiple experimental tests, this obstacle avoidance strategy demonstrated stable and efficient performance in low-speed flight scenarios, particularly in complex environments such as forests. These results confirm the feasibility of the vision-based obstacle avoidance solution using Flowdep and provide valuable insights for future drone autonomous obstacle avoidance technology, especially in complex outdoor environments.

1. Jonathas Barreto, Luciano Cajaíba , João Batista Teixeira , Lorena Nascimento, Amanda Giacomo, Nelson Barcelos, Ticiana Fettermann and Agnaldo Martins, "Drone-Monitoring: Improvingthe Detectability of Threatened Marine Megafauna," Drones 2021, 5(1), 14
2. Brendan P. Kelaher, Kim I. Monteforte, Stephen G. Morris, Thomas A. Schlacher , Duane T. March, James P. Tucker and Paul A. Butcher, “Drone-Based Assessment of Marine Megafauna off Wave-Exposed Sandy Beaches,” Remote Sens. 2023, 15(16), 4018
3. Till Kroeger, Radu Timofte, Dengxin Dai, Luc Van Gool, Computer Vision Laboratory, D-ITET, ETH Zurich VISICS/iMinds, ESAT, K.U. Leuven, “Fast Optical Flow using Dense Inverse Search,” Computer Vision – ECCV 2016,17 September 2016, pp 471–488
4. Chen-Fu Yeh, Chao-Yang Tang, Tsu-Chiao Chen, Meng-Fan Chang, Fellow, IEEE, Chih-Cheng Hsieh, Senior Member, IEEE, Ren-Shuo Liu, Member, IEEE Kea-Tiong Tang, Senior Member, IEEE, and Chung-Chuan Lo, “FlowDep – An efficient and optical-flow-based algorithm of obstacle detection for autonomous mini-vehicles,” TechRxiv. April 03, 2024.
5. Gangik Cho,Jongyun Kim and Hyondong Oh, “Vision-Based Obstacle Avoidance Strategies for MAVs Using Optical Flows in 3-D Textured Environments,” Sensors 2019, 19(11), 2523
6. Lucas, Bruce D., and Takeo Kanade. An iterative image registration technique with an application to stereo vision. Vol. 81. 1981
7. Gunnar Farnebäck, “Two-Frame Motion Estimation Based on Polynomial Expansion,” Lecture Notes in Computer Science, 01 January 2003 pp 363–370
8. E.H. Andelson and C.H. Anderson and J.R. Bergen and P.J. Burt and J.M. Ogden. "Pyramid methods in image processing". 1984.
9. M Becker, CM Dantas, WP Macedo, “Obstacle avoidance procedure for mobile robots,” ABCM Symposium series in Mechatronics, 2006
10. Yan Peng, Dong Qu, Yuxuan Zhong, Shaorong Xie, Jun Luo, Jason Gu, “The obstacle detection and obstacle avoidance algorithm based on 2-D lidar,” Proceeding of the 2015 IEEE International Conference on Information and Automation Lijiang, China, August 2015
11. Subramanian Ramasamy, Roberto Sabatini , Alessandro Gardi, Jing Liu, “LIDAR obstacle warning and avoidance system for unmanned aerial vehicle sense-and-avoid,” Aerospace Science and Technology Volume 55 , August 2016, Pages 344-358
12. Michael Santoro, Ghassan AlRegib, and Yucel Altunbasak. “Misalignment correction for depth estimation using stereoscopic 3-D cameras”. 2012 IEEE 14th International Workshop on Multimedia Signal Processing (MMSP). Sept. 2012, pp. 19–24.
13. Fadri Furrer, Michael Burri, Markus Achtelik and Roland Siegwart, “RotorS—A Modular Gazebo MAV Simulator Framework,” Robot Operating System (ROS),
Studies Comp.Intelligence Volume Number:625, 2016
14. Lorenz Meier, Dominik Honegger, Marc Pollefeys, “PX4: A node-based multithreaded open source robotics framework for deeply embedded platforms,” 2015 IEEE International Conference on Robotics and Automation (ICRA) Washington State Convention Center Seattle, Washington, May 26-30, 2015