Calibration-free Line-based Tracking for Video Augmentation

This paper presents a calibration-free line-based tracking method to augment virtual objects into a video sequence. The presented method uses a cube on the first image to calibrate camera parameters. Then the 3D model for the cube is refined by adding 3D lines to its surface. For the remaining images in the sequence, the image lines are automatically detected and matched with their 3D correspondents. The matched line pairs are used for pose estimation. The experiments show that the presented method can achieve robust performance for video augmentation without any prior knowledge of 3D model and camera parameters.