Comparison of visual descriptors for 3D reconstruction of non-rigid planar surfaces

Abstract

In the deformable surface reconstruction, one relies on local descriptors to be able to perform matching between a reference frame and a deformed frame and afterwards perform optimisation to obtain the reconstruction. The quality of surface modelling highly depends on the matching process. Therefore, I am testing the performance of different detector-descriptor pairs in order to provide hints, which of these is best suitable for 3D reconstruction of non-rigid planar surfaces.