Geometric Feature Based Vision Alignment With Partially Eroded Fiducial Mark

[[abstract]]藉由影像處理而獲知產品對位圖樣之中心點座標之電腦視覺對位是光電產品自動化製程設備中一個常常使用的對位方式，其對位的精確性對產品的品質與良率有重大的影響。工業上現行的電腦視覺對位大部分都在取得對位圖樣之影像後，利用樣板比對法(Pattern Matching)或是幾何形心法而計算求得對位圖樣之中心點座標，這些方法當遇到對位圖樣本身有部分破損缺陷時，就很容易造成其判讀的中心點座標與真實的中心點座標之間有相當之誤差。 有鑒於此，本文提出利用對位圖樣本身的幾何特徵-邊緣(Edge)與角點(Corner)資訊來辨識對位圖樣之中心點座標，只要對位圖樣中尚存有一可靠的邊緣與角點，中心點座標即可相當正確地量得，越多可靠的邊緣與角點，將可獲得更準確的中心座標與量測信心度，此方法能有效避免因對位圖樣的部分破損缺陷而影響電腦視覺對位的精確性。[[abstract]]Using computer vision to get the center point coordinate of the fiducial mark is commonly adapted for alignment in the automated optoelectronic product manufacturing equipment The accuracy of the alignment has significant influence on the quality of the products Traditionally the computer vision alignments were mainly done by pattern matching or by getting the centroid of the detected mark image Those methods may get the center cooritdinate with remarkable error when the fiducial mark was partially eroded which happened quite often in the industrial practices To overcome that shortage a geometric feature based vision alignment was proposed In this approach the center coordinate was determined by partial of the edges and/or corners of the mark that are reliable The more reliable edges and/or corners the higher accuracy and certainty of the coordinate measured will be reported The coordinate measurement can be accurate and robust until only one reliable edge and corner were left without erode