Accurate camera calibration for hand-held camera in multi-view stereo reconstruction

A flexible and accurate camera calibration algorithm with circles and concentric circle patterns is proposed for hand-held camera for multiview stereo (MVS) application, which can avoid user interaction for feature extraction in traditional planar pattern based camera calibration method. In order to automatically identify the features in image, we encode the circular patterns according to the size of the patterns. The unique code can be used to match features in images and planar patterns. We use three methods to extract the circle centers, e.g. ellipse fitting, perspective correction and circle chord midpoint approximation methods. Both simulation and real image data are utilized to validate the effectiveness of the proposed method. We use this method in multiview stereo(MVS) to reconstruct three visual hull models from image sequences. The experiments demonstrate that the concentric circle patterns have most accurate results.