基於影像之人形機器人姿態辨識系統

Image-Based Humanoid Robot Pose Recognition System

郭信宏(Sin-Hong Guo)

指導教授 ： 李祖添 ； 劉智誠

淡江大學/工學院/電機工程學系碩士班/碩士(2019年)

https://doi.org/10.6846/TKU.2019.00961

摘要

本論文設計與實現一個基於影像之人形機器人姿態辨識系統，主要分為五個部分：(1)影像擷取模組、(2)顏色辨識模組、(3)運動目標檢測模組、(4)特徵匹配模組、(5)測距模組。在Linux環境下，以機器人作業系統(Robot Operating System, ROS)建構影像辨識系統的軟體開發架構，本論文使用顏色辨識模組將影像中感興趣的顏色，以人工標記的方式標記作為特徵。使用運動目標檢測模組找出畫面中的動態物體並設為感興趣區域(Region Of Interest , ROI)，以此做為事先建立好的機器人數據圖庫特徵匹配的依據。在測距模組中利用邊界檢測與物件分割兩種方法找出物體在畫面中的資訊，並利用逆透視映射法(Inverse Perspective Mapping, IPM)計算出物體與高速攝影機間的距離。由實驗結果可得知，本論文所設計的影像辨識系統，可以同時辨識出不同種類的目標物，並成功判斷機器人目前的位置與姿態。

關鍵字

影像辨識 ； 機器人姿態 ； 特徵匹配 ； ORB特徵點 ； 邊緣檢測 ； 感興趣區域 ； 機器人作業系統

並列摘要

In this thesis, an Image-Based Humanoid Robot Pose Recognition System is designed and implemented.There are five parts:(1)model of image capture, (2)model of color recognition, (3) model of the moving target detection, (4) model of features matching, (5) model of odometry. In the Linux environment, ROS is used to establish the software development framework for the image recognition system, some interested color blocks in the image are marked as featuress by using color recognition model. Region of moving target is chosen as Region Of Interest(ROI) which is used to match the features with robtic image database by features matching model. In the odometry model, finding object data on the screen is done by two different method which are edge detection and object segmentation, and Inverse Perspective Mapping (IPM) is used to calculate the distance between the object and High-speed camera. From the experimental results, we can see that proposed image recognition system really let the different target and robot pose and position to be recognized successfully at the same time.

並列關鍵字

Image Recognition ； Robot Pose ； Feature Matching ； ORB Feature Point ； Edge Detection ； Region Of Interest (ROI) ； Robot Operating System (ROS)

參考文獻

[1] “FIRA Robot Competitions”

URL: http://www.firaworldcup.org/VisitorPages/default.aspx?itemid=3.

[2] “RoboCup Competitions” URL: https://www.robocup.org/.

[3] “Open Source Computer Vision Library, OpenCV”, URL: https://opencv.org/.

[4] Y. Wu, Y. Shan, Y. Xu, S. Wang, and Y. Zhuang, “The implementation of lane detective based on OpenCV,” IEEE Second WRI Global Congress on Intelligent Systems, Wuhan, China, 2010.