Weighted elliptic curve-fitting method with sub-pixel level in the light-probe vision measuring system

Abstract

In a light-probe vision measuring system, we first extract a rough contour of the marking points based on adaptive canny edge detection, in order to improve the extracting accuracy of the image co-ordinate of the marking points’ centre. This paper shows that the edge of rough contour may be sub-pixel extracted with the method of the grey moment. This method can be extrapolated to any position in the image plane. Finally, the image co-ordinates of marking points’ centre can be obtained with the weighted least-squares curve-fitting. In this paper, we extend the existing technique and provide a comprehensive evaluation of the proposed method. The results show that this method has higher stability precision than the exiting one. The extraction accuracy of this technique can achieve as low as 0.01 pixels. We have incorporated these ideas into the actual vision measurement of the Canny Elevator Co., Ltd.