在具有先驗已知地圖環境中無人載具的視覺定位、基於視點的規劃以及目標物追蹤

Visual Localization, Viewpoint-based Planning, and Target Tracking for Unmanned Vehicles in Known Environment with Priori Map

陳宇強(Yu-Qiang Chen)

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(2022年)

https://doi.org/10.6342/NTU202202411

摘要

本論文討論的議題為無人載具之定位、目標追蹤與路徑規劃。發展的載具包含無人地面載具(UGV)以及無人空中載具(UAV)。
在定位的部分，透過設計的彩色標記板，搭配相機模型，估計出載具的位置，並加入消失點的方法來估計角度。除此之外在自然的環境中提出了一種透過樹幹做為圓柱形的地標，並使用深度相機來定位的方法。可以在已經建立好地圖的環境中，彌補GPS等其他感測器的誤差。使用卡爾曼濾波器，達成感測器融合以及沒有標記時的位姿預測。
在追蹤的部分，透過標記板的方式，抓取目標的位置，並跟著目標移動。在這部分，提出的方法包含了避撞以及任務分配的功能，可以應用在多台載具及多個目標的情境。並且面對一個載具需要追蹤多個目標的議題，加入了視點規劃的演算法。透過計算每個目標相對於相機視野中的位置，評分一個相機與多個目標的視覺距離。最後找到與想要追蹤的目標視覺距離最短的相機位置。
搭配視點規劃的演算法，將任務擴展到軌跡規劃。在UGV的情境中，透過地標作為定位的參照物。提升看到地標的數量與品質，來提高定位的效果，本論文使用RRT的演算法來規劃路徑，將路徑的視覺分數加入考慮，規劃一個針對視覺定位有更好的觀測的路徑。
在本研究中使用不同階段的實驗以及完整的模擬環境，來驗證提出的方法。

關鍵字

無人載具導航 ； 視覺定位 ； 地標定位 ； 目標物追蹤 ； 視點規劃 ； 割草機

並列摘要

The topic of this thesis is the localization, target tracking and path planning of unmanned vehicles. The vehicles developed include unmanned ground vehicles (UGV) and unmanned air vehicles (UAV).
In the part of localization, the position of the vehicle is estimated by the designed color marker board with the camera model, and the angle is estimated by adding the vanishing point method. In addition, it is proposed a method of using tree trunks as cylinder landmarks and depth camera for localization in the natural environment. It is possible to correct the error of other sensors such as GPS in the environment where the map has been established. Using a Kalman filter, it is possible to achieve sensor fusion and position prediction when no marker is shown.
In the tracking part, the position of the target is captured by the marker board and follows the target movement. In this part, the proposed method includes collision avoidance and task assignment functions, which can be applied to multiple vehicles and multiple targets. In addition, the algorithm of viewpoint planning is added to face the problem that a vehicle needs to track multiple targets. By calculating the position of each target relative to the camera field of view, the visual distance between a camera and multiple targets is scored. Finally, the camera position with the shortest visual distance from the target to be tracked is found.
With the viewpoint planning algorithm, the task is extended to path planning. In the scenario of UGV, landmarks are used as the reference for localization. This thesis uses the RRT algorithm to plan the path, taking into account the visual score of the path, and planning a path with better observation for visual localization.
In this study, different stages of experiments and a complete simulation environment are used to validate the proposed method.

並列關鍵字

Unmanned Vehicle Navigation ； Visual Localization ； Landmark Localization ； Target Tracking ； Viewpoint Planning ； Mower

參考文獻

Mo Shan, Yingcai Bi, Hailong Qin, Jiaxin Li, Zhi Gao, Feng Lin, and Ben M. Chen, “A brief survey of visual odometry for micro aerial vehicles,” in Proceedings of IECON 2016 - 42nd Annual Conference of the IEEE Industrial Electronics Society, Florence, Italy, pp. 6049–6054, Oct. 2016.

David Gonzalez, Joshue Perez, Vicente Milanes, and Fawzi Nashashibi, “A Review of Motion Planning Techniques for Automated Vehicles,” IEEE Transactions on Intelligent Transportation Systems, Vol. 17, No. 4, pp. 1135–1145, Apr. 2016.

Andy Couturier and Moulay A. Akhloufi, “A review on absolute visual localization for UAV,” Robotics and Autonomous Systems, Vol. 135, p. 103666, Jan. 2021.

Yuanchang Liu and Richard Bucknall, “A survey of formation control and motion planning of multiple unmanned vehicles,” Robotica, Vol. 36, No. 7, pp. 1019–1047, Jul. 2018.

Julius A. Marshall, Wei Sun, and Andrea L’Afflitto, “A survey of guidance, navigation, and control systems for autonomous multi-rotor small unmanned aerial systems,” Annual Reviews in Control, Vol. 52, pp. 390–427, 2021.