現今TFT-LCD面板玻璃傳送時，玻璃破裂檢測大部分使用雷射感應器偵測其角落或邊緣。若安裝於角落則為偵測範圍過小，若安裝於邊緣則受限速度改變時需微調偵測時間，少部分使用影像辨識，但其單價過高，無法廣泛使用。
本研究應用機械視覺發展一玻璃破裂檢測系統，可安裝於玻璃傳送設備，偵測玻璃基板破裂，儘早發現破裂玻璃並予以移除，避免破裂玻璃污染機台造成產能損失。
為符合線上玻璃流動之線速度，每張影像需於62.5ms內處理完成，使用AMD Athlon 64 2800+ 1.8GHZ CPU、記憶體為1G之桌上型電腦，正常玻璃為49 ms判斷完畢，可符合生產線上檢測需求。

Currently, most of TFT-LCD glass crack detection can be implemented by using Laser sensor during transferring. However, the detected range would be too small if the sensor is installed at the corner; also the sensor’s capability would be limited by the transfer speed. Different transfer speeds require the system to fine-turn the detection time. Few of detection are accomplished by using image recognition techniques which price are too high to be widely used.

In this study, we use machine vision technique to develop a glass crack detection system. It can be installed on transfer equipment and used to detect defect glasses; once the defect being detected, it will be removed to avoid causing damage of production line and loss of production capacity.

In order to fit the speed of glass transferring in a factory, each image must be judged within 62.5ms. A desktop computer with AMD Athlon 64 2800+ 1.8Ghz CPU, and 1 GB of memory is used in our experiment. A non-crack glass could be detected within 49ms, is fast enough for online inspection.

[1]張頌榮，TFT-LCD面板之點線瑕疵自動化檢測系統，國立成功大學，製造工程研究所碩士論文，2005。
[2]陳昭彰， 運用機械視覺於液晶面板上液晶量之量測，國立成功大學,工程科學研究所碩士論文，2007。
[3]歐陽衡， 機器視覺用在封裝IC外觀瑕疵檢測之研究，國立成功大學,工程科學研究所碩士論文，2007。
[4]游順豪，應用高解析度線掃描CCD於ITO導電玻璃表面瑕疵檢測之研究，國立台灣科技大學，自動化及控制研究所碩士論文，2006。
[5]鄭睿夫，影像處理技術於載帶規範檢測之應用，國立成功大學,工程科學研究所碩士論文，2002。
[6]安乃駿，三維模型幾何特徵辨識與重建之研究，國立成功大學,機械工程研究所碩士論文，2004。
[7]繆紹綱，數位影像處理，普林斯頓國際有限公司，2005。
[8]Naccache. N.J.. Shinghal.R.. SPTA,A proposed algorithm for thinning binary patterns,IEEE Transactions on Sysytem Man and Cybernetics,Vol.SMC-14,No.3,PP.409-418,1984.
[9]Wu,W.Y.,Dominant point detection using adaptive bending value,Image and Version Computing,Vol.21,No.6,pp.517-525,2003.
[10]http://www.chimei-innolux.com/opencms/cmo/technology/面板_Size_Evolution/?__locale=zh_TW.