The objective of this work is to develop a set of interactive tools, manual and semi-automatic, to improve the quality of 3D models (acquired using laser range sensors), which the automated 3D reconstruction procedures are not able to describe. The approach followed in this work is to apply specific geometric algorithms either on the entire model or in particular regions to correct data imprecision in respect to the real word. This includes edge straightening, constrained plane flattening and other general corrections. The techniques developed are based on the automatic/semi-automatic edge detection either on the triangulated model or on the combination of registered video and range images. This is followed by the mesh modification to accommodate the edge straightening. Current work is focused on the extension of the data correction concepts to surfaces by introducing a planar flattening algorithm. These are the foundations for a further, more abstract, surfaces correction such as a plane corner. This is very useful in the case of columns, roof and floor corners, and whenever there is a 3D object having a corner between planes, very frequent in architectural environments.
The image shows plane flattening; before (a) and after (b) the orthogonal regression planar fitting.
|
|
|