- 學位論文
透明玻璃扭曲瑕疵檢測
Automated Distortion Defect Inspection of Transparent Glass
若您是本文的作者，可授權文章由華藝線上圖書館中協助推廣。
摘要
在日常生活中處處可見透明玻璃，例如：汽車上的擋風玻璃與窗戶、大門、觸控面板、液晶螢幕等多項產品。在透明玻璃中存在著透視性之扭曲瑕疵，像汽車擋風玻璃存在扭曲瑕疵時，會造成駕駛在行車時之視線所看見的外界物體產生變形、模糊等景象，易讓駕駛者在行駛中造成視覺誤判，因而造成行車上之安全顧慮。透明玻璃生產製造中常因烘烤彎曲加熱過程中，爐內溫度控制不當或是治具變形導致玻璃產生扭曲瑕疵，因此本研究提出一套透明玻璃之扭曲瑕疵檢驗方法。 本研究先利用格子線狀、黑白格子狀與圓形多點狀標準檢驗板於待測物透明玻璃之成像並擷取扭曲景像，從擷取的影像後進行影像前處理(影像增強、影像二值化/邊緣偵測)後的影像為二位元影像，再由二位元影像進行霍夫轉換(Hough Transform)搭配累積器分析找尋影像資訊峰點，再由峰點資訊進行反轉換至影像空間中並重建出新的標準影像，最後將二位元影像與重建標準影像進行影像差異比較得到扭曲瑕疵的位置，最後計算扭曲瑕疵之變形量。 本研究最主要的特色為能夠針對待測影像自動產生一張標準板的影像進行比對而判斷出扭曲瑕疵的位置處，不需要再經由人工進行標準板的比對。初步實驗針對60張(40張瑕疵影像、20張正常影像)進行小樣本實驗參數訓練。最後針對大樣本269張(150張瑕疵影像、119張正常影像)進行實驗，實驗結果可有效的判斷扭曲瑕疵是否存在與發生位置，檢出率(1-β)可達97.33%，瑕疵誤判率(α)為3.36%。
並列摘要
Transparent glass products have become necessities in our daily life and major materials for car, construction, optical and electronic industries. Since the surface distortion defects directly affect the quality of the transparent glass products, the detection of distortion defects is very important for manufacturers. In the automotive windshield manufacturing process, the furnace temperature of laminated glass controlled improperly results in skewed flaws existing on windshield in the curved baking operation. If a car windshield with distortion flaws will make object deformation and motion blur from the driver's sight easily, the drivers can cause visual misjudgment and have safety concerns on the road. This study presents the design of an automated distortion defect inspection system of car transparent glass. We will develop distortion defect detection methods for three common used standard patterns (vertical lines, grids, circle dots) in industry. In this study, a standard pattern with vertical lines displayed on a testing transparent glass is captured as testing images. First, a testing image is transformed to Hough domain to obtain the coordinates of the correct axis positions of multiple vertical lines. Through the accumulator analysis to find the peak points of the vertical lines in Hough domain, an image with new vertical lines is reconstructed from the selected peak points by taking the inverse Hough transform. Secondly, the binary testing image subtracts the binary reconstructed image to obtain a binary difference image of distortion defects. Finally, the cumulated deviation ratios of distorted segments are calculated and the offset pixel ratio of distortion segments reveals the degree of distortion levels in the image. Experimental results show that the proposed method effectively determines whether there are distortion flaws with the occurrence location, as well as distortion segment cumulative pixel ratio. The developed system achieves 97.33% distortion defect detection rate and 3.36% false alarm rate for transparent glass images.
參考文獻
[1] Adamo, F.,Attivissimo, F., Nisio, A.D, Savino, M., “A low-cost inspection system for online defects assessment in satin glass” , Measurement, Vol.42, pp. 1304-1311 (2009).
[3] Ballard, D. H., ”Generalizing the houghtransform to detect arbitrary shapes” , Pattern Recognition,Vol. 11, No.2. pp. 11 1122, (1981).
[4] Bhuvanesh, A., Ratnam, M. M., “Automatic detection of stamping defects in lead-frames using machine vision: Overcoming translational and rotational misalignment”, International Journal of Advanced Manufacturing Technology, Vol.32, pp.1201-1210, (2006).
[5] Canny, J., ”A Computational Approach to Edge Detection” , IEEE transactions on pattern analysis and machine intelligence, vol. PAMI-8, no. 6, pp. 679-698, November 1986.
[6] Chuang, M.C., ”Digital image processing image segmentation”： http://www.csie.cyut.edu.tw/~s9227609/DIP-4-segment.pdf