Abstract
This thesis addresses the problem of recovering both the camera motion and scene structure given a sequence of images captured from a stereo camera system moving through a rigid environment. Two major investigations are carried out in this thesis.In the first part, we combine a photometric cue, namely shading flow with geometric cues, to build a more robust motion estimation algorithm. This type of motion recovery algorithm, which is called flow-based method, does not require image correspondences. Using this algorithm, we show how we can disambiguate between camera translation and rotation, without requiring a wide field of view that usually comes along with a severe lens distortion. We describe the advantages and limitations of this method and present results with real images.In the second part, we formulate a new Kalman filtering approach for structure recovery. In this method, we accommodate all the available information encoding the structure. This is done by considering multiple cues, i.e., optical flow, shading flow, and stereo disparity. The aim for taking advantage of multiple cues is, first to arrive at better estimations, and second to make the system as robust as possible with respect to perturbation in any of the cues. Shading flow as one of the cues in the depth recovery/refinement algorithm plays a crucial role. We show how shading flow by being a function of the depth map gradients contributes in better estimation of the structure. We present the results based on synthetic and real imagery to evaluate the performance of the proposed algorithm both qualitatively and quantitatively. Finally, we show an application of this technique for extended scene reconstruction by combining the depth maps using the estimated homogeneous transformations.