Research On Subpixel Edge Detection Research Based On Edge Projection In The Pixel

The rapid development of modern manufacturing industry makes different parts manufacturers pay more attention to the quality of the parts,which is mostly decided by the machining accuracy of geometric features,such as length dimension,angular dimension,circularity and so on.At present,most enterprises adopt machine vision technology to detect and control these characteristic parameters.Accurate image edges detection plays an important role in this technology.In recent years,a novel subpixel edge detection method based on edge projection has been presented by Agustín et al.This method first works out the edge projection curve in a pixel with the information of local edge area and then takes its point of intersection as a subpixel edge point.It has the advantage of low complexity,high accuracy and good stability.However,it needs to be improved on some computational processes.This article first does a deep research on the principle of that method,then concretely analyze its computational process,and then improve it as the following:(1)The old method takes intercept point as a subpixel edge point.However,the intercept point is not an optimum point to describe the subpixel location of the edge in practice.In this article,we have adopted the point of intersection between the zeroaxial line whose direction vector is the edge gradient and the edge projection curve as a subpixel edge point.In addition to that,another two points on the edge projection curve are also taken as the subpixel edge points whose center of symmetry is the above point;(2)The old method uses the line to approximate the edge projection in a neighbourhood of the edge point.This approximation error can't be ignored in the detection whose testing targets is like micro-bearing with bigger edge curvature.According to this condition,this article directly treats it as a circle,then spreads it into 2 order Taylor formula,and then computes the coefficients of each term to acquire the subpixel location of the edge.In addition,this article has also researched the different operators which belong to the methods of classical operator and optimal operator,and verify them with an experiment.We also analyze the distribution type of internal noise of the image and propose a method to eliminate the external noise coming from industry field interference based on morphology.All researches we did are aiming to increase the accuracy and speed of the subpixel edge detection.At last,some experiments were done on vehicle engine air intake manifolds and micro-bearing.At the same time,we also discuss the methods for selecting experimental equipment and capturing images.The results show that our method has higher accuracycompared with old method and greater ability of anti-noise than the method based on surface fitting.It is suitable for the edge detection with high accuracy and vision measurement.