ATTENTION: This is a web archive! The IMS Group was split up in 2018 and does not exist anymore. Recent work of former members can be found at the VR/AR Group and the Computer Vision Group.

Interactive Media Systems, TU Wien

Effiziente Tiefenpropagierung in Videos mit GPU-Unterstützung

Thesis by Manuel Ivancsics

Supervision by Margrit Gelautz and Nicole Brosch

Abstract

The commercial availability of stereoscopic displays increases the demand for 3D content. Therefore, fast and robust tools for the generation of such content are of academic and commercial interest. An avenue to generating 3D content is to convert existing 2D content to 3D using specialized software. In this regard, the effort for annotation by conversion specialists and overall computation time are two crucial factors. To this end, this thesis presents an optimized semi-automatic algorithm for efficient 2D-to-3D video conversion that is based on an existing conversion algorithm. In the existing algorithm, the specialist provides sparse depth information. This depth is then propagated to the specified video clip, using segmentation and filtering techniques. The primary goal of this thesis is to reduce the computation time of the existing 2D-to-3D conversion algorithm, while maintaining the quality of its results. To achieve this goal, the most computationally expensive parts of the algorithm are re-implemented on the GPU, where they are processed in parallel. The optimization of the existing conversion algorithm also includes the optimization of the segmentation and filtering stages. The limited capacity of the GPU´s onboard memory places limits on the parallel execution of large data sets like videos. This thesis solves this problem by splitting long videos into smaller sequences and processing them sequentially while incorporating methods for handling borders between the sequences. The evaluation shows that the proposed, optimized 2D-to-3D conversion algorithm is capable of generating high-quality results, while significantly reducing execution time compared to the original, un-optimized algorithm. Additional comparisons show that the proposed optimized conversion algorithm significantly outperformed two related conversion algorithms, in terms of both quality and execution time.

Reference

M. Ivancsics: "Effiziente Tiefenpropagierung in Videos mit GPU-Unterstützung"; Supervisor: M. Gelautz, N. Brosch; Institut für Softwaretechnik und interaktive Systeme, 2014; final examination: 01-17-2014.

BibTeX

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.