基於K-means分群法全域影像定位之擬剛體編隊控制

Global Vision Positioning System based on K-means Clustering for Pseudo-rigid Formation Control

吳至原(Chih-Yuan Wu)

指導教授 ： 王立昇

國立臺灣大學/工學院/應用力學研究所/碩士(2017年)

https://doi.org/10.6342/NTU201703287

摘要

本文研究主旨為強化用於擬剛體編隊控制之載具定位與姿態識別系統，由一台高處攝影機進行影像補捉及載具監控，提供一個全域座標系統，結合影像處理與K-means分群法，使擬剛體隊形系統掌握各載具的絕對位置、姿態，並能在複雜的環境下適應及辨識出載具，將原本獨立的各載具編成擬剛體隊形後搭配路徑規劃演算法，實際整合軟硬體並進行實驗。
我們採用擬剛體編隊法設計多載具的運動隊形，將擬剛體的形變理論用在編隊設計上，使多載具隊形的變化由一組空間齊性形變張量決定，並容許隊形拉伸、剪變、旋轉，使隊形更能適應較複雜的環境。在路徑規劃部分以快速探索隨機樹(RRT)，配合路線平化技巧建構出隊伍中心路徑。接著設計虛擬位能函數來得到擬剛體的形變張量求得各載具之規劃路徑。
系統首先由M3006V網路攝影機偵測各載具初始座標與姿態，接著計算出個別載具與其所設定之擬剛體編隊初始化的隊形位置距離與期望的姿態夾角，回傳給主控電腦控制載具進行編隊初始化任務。初始化任務完畢後，由Kinect初步偵測環境並於主控電腦規劃路徑，於隊伍運動中以多載具協同機制控制載具間的運動以完成編隊運動之任務。

關鍵字

擬剛體 ； 編隊 ； K-means ； 路徑規劃

並列摘要

The purpose of this thesis is to design an initialization scheme for the pseudo-rigid formation control of a multi-vehicle system. The scheme combines K-means clustering method and image processing method. By these methods, we can identify and control each vehicle by integrating the path-planning algorithm, hardware and software in the experiments.
The concept adopted to design the formation of vehicles is the pseudo-rigid body theory, which is determined by a homogenous deformation tensor. It allows the formation to stretch, to shear and to rotate, making it adapt to the complex environment. In addition, the Rapidly-Exploring Random Tree (RRT) method was used with the technique of route adjustments to construct the motion of formation center. Then, the deformation matrix is found by utilizing the method of virtual potential function, in which the routes of each vehicle are computed.
In the initialization process, firstly we use the Webcam M3006V to detect the environment and to perform the path-planning algorithm on the main control computer. The color noises in the environment can be removed through the K-means theory. Once the initialization process is done, the coordinated control algorithm for multi-vehicle system is designed for route-following situations. The sensor Kinect was used to detect the obstacle boundary, ultrasonic system was invoked to detect the distance of obstacles, Fuzzy theory was applied to adjust the parameter of the deformation matrix. Experimental results showed that the designed initialization scheme works so that the pseudo-rigid formation can be realized to attain the destination while the obstacle can be avoided

並列關鍵字

Pseudo-Rigid Body ； Formation ； K-means ； Path-Planning

參考文獻

[1] S. L. Veherencamp, “Individual, kin, and group selection,” in Handbook of Behavioural Neurobiology, Vol. 3, Social Behavior and Communication, 1987.

[2] J. M. Cullen, E. Shaw & H. A. Baldwin, “Methods for Measuring the Three-Dimensional Structure of Fish Schools,” Animal Behavior, Vol. 13, pp. 534-543, 1965.

[4] R. G. Brown & J. S. Jennings, “A Pusher/Steerer Model for Strongly Cooperative Mobile Robot Manipulation,” IEEE Int’l Conf. Robots and Systems, Vol. 3, pp. 562-568, 1995.

[5] J. Huang, S. M. Farritor, A. Qadi & S. Goddard, “Localization and Follow-the Leader Control of a Heterogeneous Group of Mobile Robots,” IEEE Trans. Mechatronics, Vol. 11, No. 2, Apr. 2006.

[6] P. Misra & P. Enge, Global Positioning System, Ganga-Jamuna, Lincoln, MA, 2006