Abstract: An efficient feature-preserving and parameterization-free projection method for geometry reconstruction is proposed.Firstly, a local optimal projection operator incorporating bilateral filter is proposed, which takes both spatial information and geometric feature information into consideration and preserve well the geometric features in data.Then, an octree subdivision technique is applied to accelerate the algorithm, so that processing large-scale point cloud is possible.Finally a method to calculate the adaptive local support size for robust projection operator is provided.The experimental results show that the proposed algorithm is efficient and robust to reconstruct geometry and preserve features well.