Cross-Scale Cost Aggregation and PatchMatch: An Improved Dense Stereo Matching Method
Felicitas Höbelt
(Prof. Dr.-Ing. Volker Rodehorst, Prof. Dr. Charles Wüthrich)
Automatic, high-precision depth sensing is an essential task for many applications such as surface reconstruction and autonomous navigation of cars, robots or unmanned aircraft systems. However, automatic 3D-scene reconstruction from two images is not trivial. A scene point that is projected into two images can be recovered by triangulation only, if the positions of the corresponding two projections in the two images are known. For a dense reconstruction, correspondences for each pixel need to be known. This Master’s thesis merges the ideas of two publications: the SuperPixel-based PatchMatch filter framework and the cross-scale regulated cost aggregation. Additionally, by replacing a component of the cost function, the proposed method is further enhanced. The ranking estimations show that local methods still have a chance at competing with the qualitative results of global approaches.