Euclidean3-D Reconstruction From An Image Sequence

Reconstructing the3-D dimensional structure from2-D images is one of key issues in computer vision. In order to solve this problem, many approaches have been proposed in recent years. Because of requiring less constraints and pre-acquired information,3-D reconstruction based on image sequence become the main method to recover the structure information from images among these approaches. However, the traditional3-D reconstruction method based on image sequence can only realize the3-D structure in projective scale or affine scale, which will lead to ambiguity scale in the recovered structure. Therefore, we present a Euclidean3-D reconstruction method to eliminate the ambiguity and to recover the true scale of the scene in the world.In order to bring the Euclidean scale into the3-D reconstruction procedure, we measure a few nature features’3-D coordinates in the first frame of the image sequence. According to the property of the PnP algorithm, we use six2D-3D point pairs to estimate unique solution of the camera pose and position parameters in the first frame. Then by using the nature feature tracking method proposed in this article, we estimate the camera pose and position parameters from the second frame to the last frame in the image sequence iteratively. Meanwhile, we reconstruct the nature features’3D coordinates by employing the estimated camera pose and position parameters. Finally, we optimize the back-projection error function to make the whole estimated data more accurately. Because of inducting the Euclidean scale in the first frame, we can acquire the true camera pose and position parameters and nature features’3-D positions in the world coordinate. Proved by the experiments in this article, our Euclidean3-D reconstruction method demonstrates high robustness in the whole reconstruction process and accurate estimated3-D positions of the nature features.