Sub-pixel Corner Detection Algorithm

Image corner detection is one of the basic image processing tasks.In the field of digital image processing,corner has been a very intuitive and important local feature.So it plays an important role in grasping the outline of the target characteristics.The characteristics of the corner have the properties of less calculation,simple match and rotation invariance, translational invariance and scaling invariance.So corner detection widely applied in optical flow estimation,motion estimation,target pursuit,shape analysis and camera calibration.It has important value in 3D reconstruction and vision positioning and measuring.There is no clear mathematical definition for corner,but it is generally agreed that corner is the two-dimensional image of the point dramatically changes in brightness or image edge curve maximum curvature points.Corner is the picture of an important local features,it plays an important role in the field of computer vision such as the three-dimensional scene reconstruction,motion estimation,target tracking,target recognition,image registration and matching.Based on digital image,this paper makes a depth study in the sub-pixel corner detection.A sub-pixel corner detection algorithm which combined classic corner detection algorithm and curve fitting sub-pixel edge detection algorithm is proposed.This paper mainly researched in four aspects:1.As corner detection is an area of digital image measuring,so the factors that influenced image measurement accuracy also played the same role in corner detection.The sources of these errors are mainly as the factors in hardware,environment,software and so on. So we firstly analysis how these errors influenced the image measurement system precision and how to eliminated or reduced these errors.2.We need a pixel level algorithm to provided initial value for the sub-pixel corner detection.So we mainly introduced three kinds of pixel level corner detection algorithm namely Moravec corner detection algorithm,Harris corner detection algorithm and SUSAN corner detection algorithm.We chose the most efficient and exact algorithm through the experimental comparison.Meanwhile,this provide guarantee for the sub-pixel corner detection.3.We need a kind of sub-pixel corner detection algorithm to detect more exact position of corner,and then we can know the edge function.So in this paper,we analysis the working principle of various types of sub-pixel edge position operator,such as two-dimensional gray-level moment algorithm and Gaussian fitting algorithm.Finally we selected the best sub-pixel corner detection algorithm through the computer simulation experiment.4.Among sub-pixel corner detection algorithms,the improved Harris algorithm which can reach sub-pixel access to a wide range of applications.So we compared our algorithm with the improved Harris algorithm through experiment.We applied the results of our algorithm and the improved Harris algorithm to Zhang's calibration algorithm respectively,and then we compared the residual error.The results showed that our algorithm get higher accuracy.This paper is divided into six chapters.In chapter one,we summarized the concepts in corner detection and the existing algorithm,introduced the background and purpose of corner detection.Then we elucidated the main contents in this paper.In chapter two,we analyzed the factors that influenced the accuracy of image measurement system and proposed the methods to eliminate or to reduce this error factors.This provided theoretical support to the further research.In chapter three,three kinds of corner detection algorithm have been introduced,and then we compared three corner detection algorithms through experiments. Respectively,this chapter describes the outline of the former two types of corner detection method and focuses on the theory and methods of implementation of the third category. Through the comparison of three typical algorithms in the third type of corner detection algorithm,we selected Harris corner detection algorithm as the method to detect the corner coordinates of the calibration board which as the initial value of the sub-pixel corner detection.In chapter four,we introduced commonly used sub-pixel corner detection algorithms,we selected the most accurate algorithm to apply in this paper.In this paper,we get the sub-pixel corner coordinates through solve the intersection of the straight-line that constitutes the corner edge.Therefore,in this chapter,the concepts of edge detection have been introduced firstly.Then we introduced two kinds of typical pixel edge detected algorithm,namely Sobel edge detected algorithm and Canny edge detected algorithm. Sub-pixel edge positioning technology need the initial value which provided by pixel edge detection algorithm.Using this initial value,we can get the accurate sub-pixel edge position. There are lots kinds of edge detection algorithm for the two-dimensional digital image.This chapter mainly introduced two-dimensional gray-level moment sub-pixel edge detection algorithm and the Gaussian fitting sub-pixel edge detection algorithm.Both of them have advantages and disadvantages,through the comparison of experiments,we chose the Gaussian fitting sub-pixel edge detection algorithm as our algorithm.In chapter five,we compared the performance of our algorithms to the improved Harris algorithm through experiments.We concretely introduced the working principle and the realization of the algorithm which we using.Then we detected the corner of the calibration board which using the both detection algorithm.The results of the experiment showed that the algorithm which we used has better accuracy than the improved Harris algorithm.In chapter six,we summarized the results of this paper,meanwhile we carried on the forecast to this paper insufficiency as well as the next step work.