Two 2D-images, recorded from a pair of parallel cameras, have a slightly different perspective because of their horizontal displacement. This can be used to compute the pixel´s distance from the camera, which is inversely proportional to the pixel´s depth. The depth of each pixel can be represented as a gray-scale image, called disparity map. Disparity maps have become increasingly important with the development of autostereoscopic displays, which use a left image and the corresponding disparity map as input for generating 3D content with novel viewpoints. In this thesis, an automatic stereo conversion procedure for videos is proposed. First, a shot detection algorithm is presented which combines two methods: Color-Histogram and Edge Change Ratio. The experimental results are evaluated to detect the optimal parameter set for the best algorithm performance. Now, knowing the shot boundaries, the minimal and maximal disparity for every shot can be computed. From a collection of self-made stereo records and professional records, three diverse video test sets were prepared. Each sequence was captured with a different stereo-setup ranging from two commercial user cameras over a middle-class stereo camera to a professional studio camera. A comparison of these setups is presented as well as the analysis concerning the issue, if the different setups had any effect on the test results. The output of the automatic shot detection and disparity range detection is verified by manually determined values. Finally the results are evaluated and summed up.
V. Nake: "Integration von automatischer Shoterkennung und Disparitätsabschätzung in ein Stereo Matching Framework"; Supervisor: M. Gelautz, F. Seitner; Fakultät für Informatik der Technischen Universität Wien, 2012.
Click into the text area and press Ctrl+A/Ctrl+C or ⌘+A/⌘+C to copy the BibTeX into your clipboard… or download the BibTeX.