Intelligent Spatial Object Digitization based on Stereo Images

W. Li (France, Germany), M. Bhler, R. Schtze (Germany), F.S. Marzani (France), F. Boochs (Germany), and Y. Voisin (France)


close range photogrammetry, 3D surface reconstruction, meshes, pattern projection, surface curvature.


In computer vision, active stereoscopic systems are widely used for 3D surface reconstruction of real objects. Many structured light approaches have been shown in the literature. However, not anyone adapted the processing to the geometrical characteristics of the surface. Instead, the sensing process covers uniformly the entire object in order to obtain a very dense 3D point cloud. A further time-consuming mesh simplification task may therefore be necessary to simplify the manipulation of the 3D model. This can be avoided, if the sensing process is already as intelligent as necessary to generate only those points needed to optimally describe the surface. In our development we show a solution, which is based on an iterative process adapting the point distribution to the spatial structure of the object. Starting from an initial sparse point pattern a refinement process is guided which analyses the object based on curvature measurements allowing to detect areas of greater morphological variation, where further measurements have to be introduced. Thus, the acquired 3D model is already optimised during the acquisition process. Numerous experiments showed that compared to the 3D models generated from commercial systems, the loss of morphological quality is negligible, but the gain by the simplification of the model is considerable

Important Links:

Go Back