過去人類從人力、自然力來幫助生產，因科技日新月異的突飛猛進，發明了電力、 各式新物質，有了電腦協助處理許多資訊，機器人來協助做搬運及人類做不到或有能力 有限的事情，在現代工業也開始導入了自動化與智慧來輔助生產與檢驗。 在工廠生生產上往往都是以人力做操作，包括觸覺、視覺、聽覺，進行功能測試時， 像是需要聽到聲音的敘述，做出對應的動作，或按某一顆按鈕，但可能會因為人為的分 心、認知差異在沒有聽到聲音的狀況下也做出選擇，或在聽完聲音敘述後，做出錯誤的 操作造成將正常的良品認定為不良的產品，或者將不良的產品認定為正常的產品流下去 之類的誤判，使得公司遭受損失。 本論文以視覺化來識別為出發點，希望透過程式的判定，可以更精準的解決時間判 斷上的誤差，提高其精準度，並透過程式易複製增加性，已較方便的方式來配合生產狀 況使用，不會因為人員異動或培訓，造成成本開銷。研究上使用Python 程式語言搭配 scikit-image 裡的Canny 邊緣檢測、霍夫轉換…等函式庫輔助處理。透過讀取拍攝後的圖 片，鎖定並擷取只有錶面的部份，在以霍夫轉換找取指針邊緣，透過指針邊緣的資訊， 判斷當前顯示的時間是否正確，並且提供正確時間資訊。 研究中取照片 300 張，分別有手錶顏色背景為黑色、白色、深色系列，外框不同樣 式、手錶大小不一及不同時間的照片來做測試，透過程式的處理讓指針資訊能夠正確的 被抓取出來，並且做時間檢測與判定。

In the past, human helped to produce from manpower and natural resources. Due to the rapid advancement of science and technology, electricity and various new materials were invented. Computers assisted us in processing a lot of information, and robots help humans do what they cannot or have limited ability to do. In modern industry, automation and intelligence have also begun to be introduced to assist make and test. In factory production, the operation is often done by humans, including tactile, visual, and auditory. It happens that the company has a project on hand to make wearable consumer watches. There will be some problems in the production process. This problem may be caused by incoming materials or assembly need to be tested by functional testing. However, when doing functional testing, misjudgments may be caused due to human distraction and cognitive differences. For example, when performing functional tests, humans hear the instruction of the sound, do the mapping action, or press the correct button. However, sometimes because of operators' distraction or cognitive differences, they made choices even without hearing the voice, or after listening to the voice narration. When the demand suddenly increases, it may require a large number of operator, and spend time and money on personnel training, but when the demand suddenly decreases, it will cause a surplus of personnel, and there is no production and staff idle. To solve the descried problem, this thesis proposes a visual recognition system as the starting point. Through the judgment of the system, the error can be reduced and improve accuracy. The program is easy to copy and extend, and it is convenient to use in accordance with the production situation. There will be no cost due to personnel changes or the need for training. The system is developed using Python programming language, with Canny edge detection, Hough transform... and other function libraries in scikit-image to assist in processing. The captured images only contain the watch faces, which let the processing easier. This study uses the Hough transform to find the edge of the hour hand and minute hand, and then uses the information on the edge of the hands to determine whether the currently displayed time is correct or not. In the study, 300 photos were get to test. The color background of the watch is black, white, and dark series. The outer frame is different, the size of the watch is different, and the photos at different times are used for testing. The pointer information can be correctly processed through program processing it, and do time detection and judgment.

[4] Van der Walt, S., Schonberger, J. L., Nunez-Iglesias, J., Boulogne, F., Warner, J. D., Yager, N., ... Yu, T. (2014). scikit-image: image processing in Python. PeerJ, 2, e453.Howse, J. (2013). OpenCV computer vision with python. Packt Publishing Ltd.

Google Scholar