Enhancing Dense 3D Models by Multi-View Triangulation and Subsequent Illumination Estimation
Enhancing Dense 3D Models by Multi-View Triangulation and Subsequent Illumination Estimation
Masters Thesis
- Abstract:
- The reconstruction of rigid 3D objects is a well known task in current research. Especially camera based techniques are advantageous since their capturing devices are portable and cheap. In this thesis the existing Sony3D algorithm, which provides a dense 3D reconstruction of a rigid object from a sequence of 2D images, is investigated and extended. This existing algorithm relies on a two-view approach for dense reconstruction, where each reconstructed 3D point is estimated from only two images. This method has several drawbacks such as redundant and imprecisely reconstructed 3D points. In this thesis the Correspondence Chaining algorithm is developed and implemented, which enhances dense 3D models by a multi-view reconstruction, where each 3D point is estimated from multiple images. This algorithm is an extension of Sony3D and leads to an enhanced precision and reduced redundancy. The algorithm is evaluated with three different representative datasets. Sony3D with Correspondence Chaining reduces in comparison to its initial state the mean error of the reconstructed pointclouds related to ground truth data by up to 40%, whereas the root mean square error is even reduced by up to 56%. The reconstructed 3D models have a lower complexity, the file size is reduced by up to 78% and the computation time of the involved parts is decreased by up to 42%. Based on the extension to a multi-view reconstruction further estimations are possible. As an exemplary additional estimation an enhanced color estimation is developed and implemented in this thesis. Summarized, the implemented Correspondence Chaining algorithm delivers dense 3D models with improved precision and reduced redundancy in comparison to the initial state of Sony3D; the enhanced results are achieved with less memory consumption and faster computation time.