基於單眼視覺之無人飛行器三維空間之動態物定位、跟隨與空中攝影

Monocular Vision-Based Unmanned Aerial Vehicle Autonomous Flight for 3D Tracking, Following and Aerial Videography of Moving Target

張哲誠(Che-Cheng Chang)

指導教授 ： 連豊力

國立臺灣大學/電機資訊學院/電機工程學研究所/碩士(2019年)

https://doi.org/10.6342/NTU201902841

摘要

本篇論文為無人飛行器之動態物追蹤、跟隨與空中攝影提出一個完整的系統架構。在此架構中，無人飛行器倚賴單眼視覺，並將影像資訊回傳至遠端電腦，達成追蹤、定位與跟隨動態物的任務。本篇論文主要的貢獻在於，不需仰賴額外的動作捕捉系統，即可在任意環境同時定位無人飛行器並且定位地面上之動態物，以及實時規劃無人飛行器的追蹤軌跡，其考慮無人飛行器之動力學與攝影視角的優化。
其中有五個平行運作的程序來負責不同的功能。ORB-SLAM程序負責提供飛行器的自體定位與重建真實尺度的三維空間資訊。由於單眼視覺SLAM系統缺乏真實尺度資訊，因此本系統使用在一個超聲波距離感測器，估測出單眼影像系統的真實尺度。動態物姿態估測程序負責在已重建的三維空間中，倚賴單眼影像與飛行器之姿態並定位動態物。在這個程序中，首先使用Kernelized Correlation Filter抓取待被追蹤的物體，並透過物體的特徵點擷取搭配特徵點投影之影像平面的幾何關係，求得被追蹤物在三維空間的幾何座標。軌跡規劃程序負責根據動態物之位姿，計算出飛行器應抵達的攝影位置與相對動態物移動方向的特定角度，並規劃平滑且符合無人機動態行為的軌跡。此程序規劃以多項式為基底的平滑軌跡，確保位置的四階微分函數具有連續性，並且隨時控制被追蹤物在影像中的行位置保持在影像平面中心。姿態控制器程序根據軌跡規劃的結果，控制飛行器以抵達該指定位置。
最後，本論文展示了多項模擬以及實驗結果，以證明此系統的可行性以及性能。

關鍵字

航空追蹤系統 ； 單眼視覺同步定位與地圖建構 ； 影像伺服 ； 即時系統 ； 路徑規劃 ； 空中攝影

並列摘要

In this thesis, the autonomous flight system of a quadrotor is proposed for moving object tracking, following and aerial photography. In this system, the quadrotor relies on monocular vision and transmits RGB image back to remote computer to perform the task of tracking, locating, and following moving target. Our key contributions include localization of unmanned aerial vehicle and moving target on the ground in any environment without relying on motion capture system as well as planning algorithm which considers dynamics of the unmanned aerial vehicle and optimization of photographic perspective.
There are five parallel programs that are responsible for different functions. The ORB-SLAM program is responsible for providing localization of the quadrotor and reconstruction of real-scale 3D spatial information. Since monocular SLAM system lacks real-scale information, the system uses an ultrasonic range sensor to estimate the true scale of monocular SLAM system. The dynamic target pose estimation program is responsible for locating the dynamic objects in the reconstructed three-dimensional space using onboard monocular image and the attitude of the aircraft. In this program, the Kernelized Correlation Filter is utilized to track target on image plane, and the positions of tracked target in the three-dimensional coordinate are obtained through the feature points of the target in image plane. The trajectory planning program is responsible for setpoints calculating, planning a trajectory that conforms to dynamic behavior of the quadrotor and visibility optimization on image plane. This program plans a smooth polynomial-based trajectory, ensuring continuous property of forth derivative of position, and controls column position of the tracked target at the center of image plane at the same time. The attitude controller program controls the quadrotor to arrive at the designated position based on the results of the trajectory planning.
Finally, the analysis in both simulations and experiments is provided to demonstrate feasibility and performance of proposed system.

並列關鍵字

Aerial tracking systems ； monocular SLAM system ； visual servoing ； real-time system ； path planning ； aerial photography

參考文獻

[1: Jordan et al. 2018] Sophie Jordan, Julian Moore, Sierra Hovet, John Box, Jason Perry, Kevin Kirsche, Dexter Lewis and Zion Tsz Ho Tse, “State-of-the-art technologies for UAV inspections,” IET Radar, Sonar & Navigation, Vol. 12, No. 2, pp. 151-164, Feb. 2018.

[2: Vempati et al. 2017] Anurag Sai Vempati, Igor Gilitschenski, Juan Nieto, Paul Beardsley and Roland Siegwart, “Onboard Real-time Dense Reconstruction of Large-scale Environments for UAV,” in Proceedings of IEEE/RSJ International Conference on Intelligent Robots and Systems, Vancouver, BC, Canada, pp. 3479-3486, Sep. 24-28, 2017.

[3: Tomic et al. 2012] Teodor Tomic, Korbinian Schmid, Philipp Lutz, Andreas Domel, Michael Kassecker, Elmar Mair, Iris Lynne Grixa, Felix Ruess, Michael Suppa and Darius Burschka, “Toward a Fully Autonomous UAV: Research Platform for Indoor and Outdoor Urban Search and Rescue,” IEEE Robotics & Automation Magazine, Vol. 19, No. 3, pp. 46-56, Sept. 2012.

[4: Thomas et al. 2016] Justin Thomas, Giuseppe Loianno, Kostas Daniilidis and Vijay Kumar, “Visual Servoing of Quadrotors for Perching by Hanging From Cylindrical Objects,” IEEE Robotics and Automation Letters, Vol. 1, No. 1, pp. 57-64, Jan. 2016.

[5: Thomas et al. 2017] Justin Thomas, Jake Welde, Giuseppe Loianno, Kostas Daniilidis and Vijay Kumar, “Autonomous Flight for Detection, Localization, and Tracking of Moving Targets With a Small Quadrotor,” IEEE Robotics and Automation Letters, Vol. 2, No. 3, pp. 1762-1769, May 2017.