Research On Automatic Camera Calibration Based On Circular Control Points

Camera calibration in the context of machine vision is the process of determining the relationship between 2D image coordinates and 3D reference coordinates. During the geometrical camera calibration, an image of calibration objects, whose shapes and dimensions are prior known, is acquired. Because 3D calibration object is easy to manufacture with low-cost and could provide monview noncoplanar control points, it is widely used in camera calibration. There are different types of control points painted on the surface of 3D calibration object, how to obtain high precision camera parameters with proper reference objects is a main problem in the research field of camera calibration. In this thesis, we present a camera calibration method based on circular points. It comprises the following contents:(1) A new easy camera calibration based on DLT model is presented. It accounts for major camera distortion and estimates the perspective transformation matrix along with distortion parameters through the iterative algorithm. And then we describe how to reconstruction the 3D coordinates of point using the result of the previous procedure. The validity and performance of our calibration procedure are tested with real images with different distortion models. The results consistently show significant improvements over the classic approach of Tsai.(2) Circular points are used to deal with the problem of choosing a proper threshold value in the step of detecting position of the control points in the images. But they introduce an asymmetric perspective bias needing corrected. The mathematical formulations of the correction terms based on DLT model are deduced from space analytic geometry theory and projection model. Synthetic and real data have been used to test the validity and performance of our method. The relationships between bias error and the parameters of the circle are reported. The experimental results show that the calibration error after correcting asymmetric projection is reduced.(3) It is necessary to order 2D coordinates of circular points in correspondence with the original 3D position of each center of circle. In previous, this step of finding correspondences is often performed manually. Our goal is to avoid this in order to make it fully automatic. Experiments show that our method is robust for pattern images under uneven illumination or in different scenes.