應用機器視覺於泡棉直徑檢測系統之開發

在競爭激烈的自動化產業中，品管屬於最重要也是最耗費人力的工作。低檢測速率、不精確、容易出錯、及大量生產時無法維持穩定的品質是傳統人工檢測的缺點。有鑑於此，大多數製造業開始採用自動化光學檢測技術，以提高檢測速率、改善檢測品質。本論文的目標是發展一套精確快速以機器視覺為基礎的直徑檢測系統。研究方法是利用CCD擷取泡棉影像，所擷取的影像是灰階影像，由於灰階影像包含太多影像資訊，所以將泡棉影像經過二值化處理，以減化處理資料量及加快檢測速率，再將影像經過Sobel邊緣偵測法處理，以獲得泡棉影像之邊緣，由計算泡棉影像的邊緣像素點之斜率的方法進行泡棉直徑檢測。另外與對灰階影像量測直徑及對經過BEA邊緣偵測法處理之影像量測直徑做比較。對於泡棉直徑之檢測研究，為了避免泡棉在檢測時產生晃動，進而影響檢測結果，所以設計一機構以限制泡棉的晃動。本研究以Borland C++ Builder程式語言撰寫泡棉直徑檢測系統之人機介面及演算法。由實驗結果得知量測直徑的方法於量測硬式試件(塊規)，可以準確到0.1mm，檢測一張泡棉影像的直徑約0.23秒。

關鍵字

邊緣偵測；自動化光學檢測；機器視覺；直徑檢測

並列摘要

In the vigorously competitive automation industry, quality control is the most important and labor-consuming task. Low inspection rate, inaccuracy, error-prone, and the inability to maintain a stable inspection quality for mass production are the main disadvantages of the conventional manual inspection. Accordingly, most of the manufactures start to adopt automatic optical inspection techniques in order to increase the inspection speed, improve the inspection quality. The purpose of this thesis is to develop an accurate and rapid machine vision-based diameter inspection system. First, image is collected by using the CCD camera and the frame grabber. Due to the fact that gray-level images contain too much information about non-target object, the image is thresholded to decrease noise and background and the Sobel edge detection is used to search the edge points, then by calculating the slope of the edge points to determine diameter. In addition to measure diameter of binary image, BEA (Boundary Extraction Algorithm) is also used to search the edge points. In order to avoid the vibration while proceeding on-line inspection, a mechanism is designed to restrict the system such that the vibration is reduced. This system, including human-machine interface and all the algorithms, is written in Borland C++ Builder language. The experimental result of the precision is up to the order of 0.1 mm .It costs approximately 0.23 seconds to inspect diameter of soft plastic pipe image.