3D recovery with free hand camera motion

3D recovery from a set of images has been one of the main efforts in computer vision. In the last twenty years, a large number of approaches have been developed, which have been applied to different areas. In this work, a 3D recovery system for a rigid scene is proposed, based on the acquisition of an image sequence taken with a camera undergoing free motion. The sequence of images must have a set of features which are matched all along the sequence. The system does not require any knowledge about camera position or any previous model about the 3D scene, it is only required the camera calibration matrix. Once the feature set and the camera calibration matrix is known, the 3D recovery process can be done. First, based on the epipolar restriction for two views, an initial reconstruction with the first and the last view is calculated. After two view reconstruction, an improvement can be carried out with a factorization using some of the other views. Execution time depends on features number, the number of images and the position of the features in the images. In our tests, the average time execution was less than two minutes.