Reconstruction tridimensionnelle en vision par ordinateur : cas des cameras non etalonnees
Boubakeur Boufama · HAL (Le Centre pour la Communication Scientifique Directe) · 1994
This thesis mainly concerns 3D reconstruction from images taken with uncalibrated cameras. Two other topics, very close to 3D reconstruction, were treated here epipolar geometry computation and points tracking in images. We consider the general case of an image sequence obtained with one or more cameras with unknown characteristics. When only image points are used, the only kind of reconstruction that can be obtained is projective. First we propose a method to compute the projective reconstruction, then we show how to get euclidian reconstruction from the projective one. This is done by using euclidian constraints arising from a priori knowledge about the scene. Finally, we propose a method for euclidian reconstruction when an approximation of the intrinsic parameters of the camera are known. In this method we used a minimal number of parameters, resulting in stable computations. All the methods presented in this thesis are validated with real and simulated images.