本論文提出一套具有高重複性的邊緣偵測與方向量測演算法,特別針對於三維的粗糙表面。我們提出的方法主要分成兩個部分。一、透過K-Means 技巧輔助來尋找合理的物體邊界。二、利用離群值檢測技巧,將前一部分所提取出的物體邊界進行最後篩選,最後將合理的邊界進行線性回歸,計算出邊緣的長度與角度。需要克服的困難包括崎嶇的金屬表面、低對比邊界、短邊界、假邊界、高重複性 ...等等。本論文的資料來自於源台精密所提供,資料中所涵蓋多種現實精密量測業界會遇到的情況,包含崎嶇的金屬表面、低對比邊界、短邊界、假邊界 ... 等等。本論文提出的方法是已經經過業界嚴格的考驗,同時考慮了演算法的效能以及重複性的檢驗。
This thesis proposes a set of edge detection and direction measurement algorithms with high repeatability, especially for three-dimensional rough surfaces. Our proposed method is mainly divided into two parts: 1. Use the K-Means technique to assist in finding a reasonable object boundary. 2. Using the outlier detection technique, the boundary of the object extracted in the previous part is finally screened, and finally the reasonable boundary is linearly regressed to calculate the length and angle of the edge. The testing cases in this thesis is provided by ARCS Precision Technology. The testing cases indeed covers various situations that the real precision measurement industry will encounter: rugged metal surfaces, low-contrast boundaries, short boundaries, false boundaries, and so on. Our proposed method has been rigorously tested in the industry, while considering the effectiveness of the algorithm and the repeatability of the testing data.