機器視覺應用於車載鏡頭鏡片表面瑕疵之檢測

ITO Resin/AgNPs Thin Film for Thermal Insulation Performance Improvement of Low-Emissivity Glass

本文主要針對車載鏡頭的塑膠鏡片利用影像處理技術做表面瑕疵的自動檢測，塑膠鏡片在製程中容易產生刮痕、汙垢、氣孔等的瑕疵，這些瑕疵很難用人工目視檢測出來。首先，建立機械視覺的實驗平台，包括個人電腦、Basler Gige乙太網路、遠心鏡頭、CCD照相機、LED環形燈光和定電流調光器。然後發展影像處理方法以檢測鏡片瑕疵。在影像定位方面，本文利用模板匹配和二維計量演算法以找出鏡片所在的位置。在瑕疵檢測方面，先以高斯拉普拉斯濾波器強化瑕疵的邊緣形狀，以動態閥值二值化分割出瑕疵的影像。再以形態學的膨脹和侵蝕法連接破碎瑕疵圖像，以斷開與連通域的方法標示瑕疵的位置，然後透過特徵篩選對瑕疵面積大小進行計算。最後，透過實際的塑膠鏡片的檢測實驗來驗證本文所提方法的有效性，瑕疵辨識成功率可達97%。

This paper examines the image processing technology to automatically detect surface defects happening to the vehicle plastic lens. Defects usually happen during the lens manufacturing process such as scratches, dirt, and pores, and these defects are difficult to detect manually. To deal with these issues, firstly, an experimental platform is established for machine vision, including a personal computer, Basler Gige Ethernet, telecentric lens, CCD camera, LED ring light and constant current dimmable driver. Image processing methods are then developed to detect lens defects. Secondly, template matching and 2D metrology algorithms are used to position the lens. Thirdly, Gaussian Laplacian filter is used to enhance the defect edge, and then dynamic binarization is used to segment the defect. Fourthly, the defect image is connected by morphological expansion and erosion method. Fifthly, the location of the defect is marked by opening and closing method, and the size of the defect area is calculated through feature extraction. Finally, the effectiveness of the method proposed in this paper is verified through the actual plastic lens detection experiment, and the defect identification success rate is up to 97%.