黃漢邦臺灣大學：機械工程學研究所吳彥勳Wu, Yen-HsunYen-HsunWu2007-11-282018-06-282007-11-282018-06-282005http://ntur.lib.ntu.edu.tw//handle/246246/61365全域視覺足球機器人系統的發展及實現必須要結合跨領域的各種技術。本篇論文主要著重於其中三個部分：彩色物件之識別和追蹤，雙輪驅動機器人之控制和多重機器人之合作。全域視覺模組需要監控整個球場的情勢，進而萃取出有意義之資訊提供策略決策機制規劃球隊策略。這些訊息的可靠度及準確性將左右戰術規劃的效率。在以色彩為基礎之影像分析辨識系統中，強韌的色彩分類扮演了極為重要的角色。適當的顏色分類大幅降低計算所需的時間，透過忽略無關緊要的背景資料，影像處理的可靠度將大大地提升。本文所發展的視覺系統採用主成分分析法 (PCA) 尋找適合色彩分類的空間，並且建立色彩分類之模型。根據此模型，色彩的分類對於光線條件稍微地差異具有一定的強韌性。按照影像模組所擷取出來的資訊，決策機制評估球場上的狀態，進而規劃畫多重機器人合作之策略。策略實行的成效決定於是否適當地分配球場上各球員扮演之角色。一個由下而上的策略產生機制，顯著地簡化策略規劃的過程，尤其是為劇烈改變的動態環境建立有效的策略。雙輪驅動機器人的動作控制則採用動態迴授線性化控制。此控制方法足以應付足球比賽中，機器人運球及得分的各種動作。本文完整地整合、發展並實現全域視覺足球機器人系統。本系統已贏得2004台灣全國賽的第一名，並且將參加2005 FIRA RoboWorld Cup比賽。Wide range of technologies are integrated, developed and implemented in the global vision soccer robot system. This thesis is involved in three of them: Color object recognition and tracking, wheeled mobile robot control, and multi-agent collaboration. The task of global vision module is to extract meaningful data for strategy decision module. These data are reliable and accurate for efficient tactics planning. Robust color classification plays a dramatic role for color-based image analysis. In addition, appropriate color classification reduces computational time and improves reliability by eliminating uninterested background information. Principal Component Analysis (PCA) is adopted to seek for a color space in which a color classification model is constructed straightforward. According to this model, slight color variation is robustly classified. Based on extracted data, the decision-making module estimates the field condition, and then plans strategies to collaborate agents. Role assignment enormously affects the efficiency of plan achievement. A bottom-up strategy generation mechanism simplifies the procedure to create strategies for highly dynamical environment. Dynamic feedback linearization control is adopted for the mobile robot’s motion control. The performance is good enough for robot to dribble the ball and score goal. The entire system has been completely implemented, and won the first place of 2004 National Competition of Soccer Robots.摘要 i Abstract ii Contents iii List of Tables vii List of Figures viii Chapter 1 Introduction 1 1.1. Motivation and Literature Survey 1 1.2. Objectives 4 1.3. Contributions 6 1.4. Organization of this Thesis 7 Chapter 2 Background Knowledge 9 2.1. Sketch of Soccer Robot Competition 9 2.2. Coordinate System Transformation 12 2.2.1. Coordinate Relation between WCS and CCS 13 2.2.2. Coordinate Relation between CCS and ICS 13 2.3. Brief Concept of Camera Calibration 14 2.3.1. Direct Linear Transformation 15 2.4. Color Space 16 2.4.1. Colorimetric Color Spaces 18 2.4.2. Subtractive Color Spaces 20 2.4.3. Nonstandard Color Space 21 2.5. Principal Component Analysis (PCA) 22 2.6. Principal Moments of Inertia of Areas 27 2.7. Mobile Robot Kinematics 30 2.8. Feedback Linearization 34 2.8.1. Input-State Linearization 34 2.8.2. Input-Output Linearization 36 Chapter 3 Color Object Recognition 39 3.1. Color Space Selection 39 3.2. Removing Background 42 3.3. Color Classification 45 3.4. Look-up-table Construction 48 3.5. Blob Aggregation 49 3.6. Object Posture Recognition 51 Chapter 4 Wheeled Robots Control 54 4.1. System Modeling 54 4.2. Control Properties 56 4.2.1. Controllability and Stabilizability at a Point 56 4.2.2. Controllability and Stabilizability about a Trajectory 57 4.2.3. Static Feedback Linearizability 58 4.3. Dynamic Feedback Linearization 60 4.4. Controller Design 62 Chapter 5 Strategy and Multi-Robot Collaboration 65 5.1. Basic Actions 65 5.1.1. Go to a Target Point 66 5.1.2. Path Planning and Path Following 66 5.1.3. Obstacle Avoidance 67 5.2. Behaviors for Different Roles 67 5.3. Role assignment 68 5.4. Collision Free Strategy 70 Chapter 6 Implementation 71 6.1. System Overview 71 6.2. Hardware 72 6.2.1. Robot Design 72 6.2.2. Wireless Communication System 77 6.2.3. Host Computer 78 6.2.4. Camera and Grabber Card 79 6.3. Camera Calibration 80 6.3.1. Distortion Elimination 80 6.3.2. Camera Position and Orientation Adjustments 83 6.3.3. Parallax Correction 84 6.4. Software 85 Chapter 7 Experimental Results 89 7.1. Experiment Setup 89 7.2. Performance of Vision Module 90 7.2.1. Resistance against Lighting Conditions 92 7.2.2. Computational Capability 97 7.3. Posture Stabilization 98 7.3.1. Simulation 98 7.3.2. Experimental Results 101 Chapter 8 Conclusions and Future works 105 8.1. Conclusions 105 8.2. Future works 106 References 108 Appendix 1134557422 bytesapplication/pdfen-US全域視覺色彩分類足球機器人Global visionColor classificationSoccer robotthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/61365/1/ntu-94-R91522831-1.pdf