俱視覺回授控制之機器人藝術繪畫應用影像分析及機器學習方法

Robot Artistic Painting with Visual Feedback Control Using Image Parsing and Machine Learning Methodologies

劉育榕(Yu-Jung Liu)

指導教授 ： 羅仁權

國立臺灣大學/電機資訊學院/電機工程學研究所/碩士(2018年)

https://doi.org/10.6342/NTU201803445

摘要

近年來，世界各地的機器人大多被用來處理工廠自動化之研究，但是，機器人技術的應用不僅限於工廠領域和產值的增加，現今越來越多的機器人研究團隊投入到教育、醫療長照、家庭服務、娛樂等領域發展，他們都朝著模仿人類行為的目標前進，以帶來更多附加價值，為了打破科技給人們較為死板的印象，我們嘗試將藝術融和到機器人當中，並給予機器人適當的創作能力，帶來與以往借由複製藝術品所做的研究有所區隔，試圖創造出獨特的作品。

本研究的目標在於將藝術創作融合機器人領域的相關技術，讓機器人能夠繪畫出具有藝術價值的獨特作品。本研究呈現兩種不同風格及手法的藝術作品，第一種著重在卡通風格的人臉肖像轉換，我們將模仿人類藝術家，運用相機作為藝術家的眼睛，捕捉人臉的五官特徵，有效分解成單個組件，再透過機器學習方法與預先收集的卡通五官資料庫做相似度比較，並選出最相近的取代原本的五官，以此來達到機器人創作的能力，接著再通過影像處理等方法降低色彩複雜度以符合卡通風格的色彩呈現，最後，機器人使用五種基本顏色以及奇異筆繪畫出具有卡通風格的作品。第二種研究我們希望能夠打破了傳統機器人只在畫紙上進行繪畫，而嘗試在不同材料的創作，我們選擇在曲面的木頭上作烙畫，利用影像處理結合機器學習方法，使用線條畫來呈現不同顏色的深淺與立體的視覺感受，最後加上我們原有的彩色繪畫功能，映射在曲面的木頭上成為獨特又新穎的烙畫藝術。

此研究充分展現了人工智慧在創造藝術價值的可能性，我們打破了機器人創造複製作品和使用單一媒介的限制，希望透過科技技術的方法能夠輔助人們學習繪畫的創作以及創新的想法。

關鍵字

服務型機器人 ； 娛樂型機器人 ； 藝術機器人 ； 機器人視覺系統 ； 人工智慧 ； 影像分析

並列摘要

In recent years, robots around the world have mostly been used to deal with factory automation research. However, the application of robotics technology is not limited to the increase of factory field and the gross of the product. Nowadays more and more robot research teams are engaged in the fields of education, medical service, family services and entertainment. They are all moving towards the goal of imitating human behavior in order to bring additional value. In order to smash the trammels of old tradition of technology, we have tried to integrate art into robots and give robots appropriate creativity. This has led to the different between researchers who study in copying artworks and attempts to create unique artworks.

The goal of this study is to integrate the art creation into the field of robotics. Allows the robot to paint unique artworks with artistic value. This study presents two kinds of artworks with different styles and techniques. The first focus on the cartoon style of face portrait conversion. We use camera as artist's eyes to imitate human behaviour that captures facial features and effectively decompose into individual components. Then using machine learning method and pre-collected cartoon facial features data to select the most similar one to replace the original. Therefore, we achieve the creativity of robot itself. After that, the robot use image processing to reduce the color complexity to match the cartoon style of color rendering. Finally, the robot uses five basic colors and sharpie marker to paint a cartoon-style portrait. The second research, we hope to break the traditional that researchers mostly paint on the drawing paper. We focuses on the creation of different materials. We choose to make a branding on the 3D surface of the wood. Using image processing and machine learning skills to generate line painting with different depth of color to create stereo feeling. Eventually, we add our colorful painting function and project on the 3D surface to create a unique and novel art of pyrography.

This research fully demonstrates the possibility of artificial intelligence in creating artistic value. We have broken the limits of robots creating duplicate artworks and using a single medium. We hoped that through technological methods, people can learn the creative and innovative ideas of the painting.

並列關鍵字

service robotics ； entertainment robotics ； art robotics ； robotic vision system ； artificial intelligence ； image parsing

參考文獻

[1] “Provisional definition of service robots english, 27th of october 2012,”

[2] “Moley robotics kitchen robot: http://www.moley.com/,”

[3] “Automatic self-cleaning litter box for cats: https://www.litter-robot.com/,”

[4] “Robot security system: http://robotsecuritysystems.com/,”

[5] J. Broekens, M. Heerink, and H. Rosendal, “Assistive social robots in elderly care: A review. gerontechnology, 8 (2), 94-103,” Zugriff am, vol. 11, p. 2012, 2009.