以幾何特徵分析為基礎的工件認識算法之開發

The Development of Workpiece Understanding Algorithms Based On Geometrical Analysis

黃世穎(Shin-Yin Huang)

指導教授 ： 蔡曜陽

國立臺灣大學/工學院/機械工程學研究所/碩士(2013年)

https://doi.org/10.6342/NTU.2013.02274

摘要

自動光學檢測(Automated Optical Inspection，簡稱AOI)系統是一種結合了精密自動化機械、光學取像系統、邏輯演算技術等核心科技的光學影像檢測系統。
在現今講求高效率、高速化的工業產線上，AOI系統不僅能改良傳統上以人力使用光學儀器進行檢測的缺點，其應用層面包括從高科技產業之研發、製造品管，以至國防、民生、醫療、環保、電力…等領域。然而現今較具代表性的AOI系統多為以模板比對法為主的檢測系統，使得若要量測工件表面的尺寸時仍得將工件以其它儀器或是依賴人工方法來進行。為了補足尺寸量測上的不足之處，本研究則開發了一結合影像處理技術與幾何推理計算的影像重建算法，在利用圖像上的各輪廓的幾何特徵來把取得圖像之資料分割成符合各特徵定義的子資料群後，再將根據子資料群的幾何性質重組成一完整之輪廓圖，而輸出的輪廓圖不僅可以做為與工件之原設計圖進行比對之素材，分類過程中的各特徵點也能透過座標計算而生成工件各輪廓的尺寸資料，使得整個系統在測得工件缺陷之餘，還能有不錯的尺寸量測能力。
為了讓系統輸出的特徵點座標具有實際的長度意義，本研究則利用了一標準塊進行長度量測的觀察實驗，並建立了一線型的像素長度比例關係式。在以此比例關係式為長度量測基礎的狀況下，本研究所建立的影像處理方法不僅能完整地對線段、孔洞、凹槽、圓角、U型槽等常見的加工特徵進行輪廓重建，對於重建後的輪廓尺寸，本算法輸出的尺寸資料在線段部分之量測誤差平均百分比約為0.808%曲線部分之量測誤差平均百分比則為2.363%，且整個量測過程能在2.8秒內完成。顯示本研究所開發之輪廓重建算法不僅能適用於大部分的機械零組件，其運作速度亦能符合線上即時檢測的需求，讓AOI系統在檢測工件表否是否有瑕疵之餘，同時告知使用者工件的加工尺寸狀況。

關鍵字

機器視覺 ； 影像處理 ； 輪廓重建

並列摘要

AOI (Automated Optical Inspection) system is a combination of precision automation machine, optical image capture system, the logical calculus technology and other core technologies optical image detection system. In today's industrial production line, AOI system can not only improve the traditional optical instruments used to detect human shortcomings, the application level is also quite extensive. But most of the AOI systems today mostly depend on template matching method so that the workpiece surface to measure the size of the workpiece to still have to rely on other instruments or manual methods. To complement the dimension measurement on the inadequacies of today’s AOI systems, this study is the development of a combination of image processing techniques and geometric reasoning image reconstruction algorithm for computing. In the use of images on the geometric characteristics of the contours of the acquired image data into line with the characteristics of each group after the definition of subfolders, and then under the sub group re-composition of a complete geometric properties of the contour map.
Furthermore, in order to make the feature point coordinates of the system output with the actual length significance, the studies carried out using a standard block length measurements observation experiment, and the establishment of the line type of relationship of the pixel length ratio. In this proportional relationship between the amount of foundation for the long position, the institute created image algorithm can complete on line, holes, grooves, rounded, U-shaped groove machined features such as common contour reconstruction, while the size measurements, but also can achieve line error 0.808%, curve error 2.363% of measurement results.

並列關鍵字

Machine Vision ； Image processing ； Contour reconstruction

參考文獻

[14] Bartholomew O.Nnaji, Tzong‐Shyan Kang, Shuchieh Yeh,Jang‐Ping Chen,

[2] D. W. Manthey, K. N. KnappII,and D. Lee, “Calibration of a laser range-finding coordinate-measuring machine”, Opt. Eng.33 (1994), p.3372–3380

[3] B. Kusnoto, C.A. Evans, “Reliability of a 3D surface laser scanner for orthodontic applications”,Am J Orthod Dentofacial Orthop, 122 (2002), p.342–348

color structured light and multi-pass dynamic programming”, 3D Data

p. 942–947