Comparison of Self-Calibration Algorithms for Moving Cameras

Abstract- English It is an important requirement for many computer vision applications that the imaging parameters of a camera used for image acquisition are known. The intrinsic and extrinsic parameters of the camera models the mapping between the scene and the images. If the values of these parameters are available, the camera is calibrated. Traditional methods for camera calibration use special patterns with known 3D structure to obtain the camera parameters. This task is hard to automatize since it requires user interaction. Self-(or auto-) calibration is an alternative and enables camera calibration on the fly while using the camera in a vision task. This thesis examines self-calibration algorithms and gives and extensive evaluation and comparison of several methods. The main focus lies on the calibration of intrinsic parameters of a freely moving camera in an arbitrary scene. No assumptions about the motion of the camera or the scene structure are made. The methods use the epipolar geometry between pairs of views with the associated fundamental matrices. Additionally, algorithms for the estimation of the relative camera motion in scene space encoded by the extrinsic parameters are introduced