Abstract |
In Dynamic Stereoscopic Vision the problem of determining the position and the motion of the depicted objects is often approached by solving the Stereoscopic and the Motion Estimation problems separately, thus not utilizing the relations between the motion and disparity fields simultaneously. This work aims at an integrated solution to the problem, by fusing the interdependencies of motion and disparity fields into a joint estimation process. Given a stereoscopic image sequence, at each time instant two dense motion fields, for the left and the right image sequences, and one dense disparity field are estimated simultaneously. The disparity field of the previous stereoscopic pair is considered as known, that is previously estimated. The estimation is the result of the minimization of a cost function which contains known equations regarding velocity and disparity fields in relation to image intensities, and which also constrains the different fields to be adaptively smooth. This minimization is achieved using an iterative multiscale relaxation algorithm based on the gradient of the cost function. The analysis includes the detection of stereoscopic occluded areas, and motion occlusions and disclosures, with a method based on confidence measures. For the determination of the disparity field of the first stereoscopic pair of the sequence, a method based on a similar algorithm is proposed. Another problem which is addressed in this thesis is the construction of intermediate views between a stereoscopic pair of cameras. The analysis uses the results of the dense disparity field estimation and stereoscopic occlusion detection of the previous methods. Experimental results on real stereoscopic image sequences are given in order to demonstrate the proposed methods.
|