Reconstruction with 3D geometric bilateral filter

In recent years, reverse engineering (RE) techniques have been developed for surface reconstruction from 3D scanned data. Typical sampling data, however, usually is large scale and contains unorganized points, thus leading to some anomalies in the reconstructed object. To improve performance and reduce processing time, Hierarchical Space Decomposition (HSD) methods can be applied. These methods are based on reducing the sampled data by replacing a set of original points in each voxel by a representative point, which is later connected in a mesh structure. This operation is analogous to smoothing with a simple low- pass filter (LPF). Unfortunately, this principle also smoothes sharp geometrical features, an effect that is not desired. The high performance results of bilateral filtering for removing noise from 2D images while preserving details motivated us to extend this filtering and apply it to 3D scan points. This paper introduces anisotropic 3D scan point filtering, which we have defined as 3D Geometric Bilateral Filtering (GBF). The proposed GBF method smoothes low curvature regions while preserving sharp geometric features, and it is robust, simple and fast.