In this paper

we describe a fast method for arbitrary surface reconstruction from 3D scattered points based on RBF (radial basis functions) implicit surface model. According to the properties of 3D data from laser scanners

such as non-uniform

large scattered and topology unknown

we emphasize the importance of data simplification and surface geometry preserving. The crucial idea is based on the difference of nearest normal vectors and covariance matrix. We transform global implicit function into local RBF models by the partition of unity method with K-d tree data structure and the weight coefficients of local model are directly derived from solving a linear system. To smooth final surface model

these local shape functions are blended together in overlapping domains. Experiment results show that the proposed approach is not only easy

fast to implement in reconstruction from large scattered point

but also with good accuracy.