A region-growing algorithm using parallel computing for surface reconstruction from unorganized points

• We develop a region-growing algorithm based on parallel computation architecture.
• The old algorithm needs to be redesigned to fit the requirement of GPU architecture.
• Our algorithm can speed up the mesh reconstruction over 10 times.
• The proposed algorithm also have the same quality compared to original algorithm.

A parallel computing region-growing algorithm for surface reconstruction from unorganized point clouds is proposed in this research. The traditional region-growing algorithm belongs to sequential process and needs to update the topology information continuously to maintain the boundaries of the growing region. This constraint becomes a bottleneck for efficiency improvement. The proposed GPU-based region-growing algorithm is to decompose the traditional sequence and re-plan specific framework for the purpose of utilizing parallel computation. Then, a graphics card with multi-processing units will be used to build triangles in the parallel computing mode. In our GPU-based reconstruction process, each sampling point is regarded as an independent seed and expands simultaneously until all surrounding patches overlap each other. Following this, the overlapping patches are removed and holes are filled by the GPU-based calculation. Finally, a complete model is created. In order to validate the algorithm proposed, the unorganized point cloud was obtained by a 3D scanner and then reconstructed using the parallel computing region-growing algorithm. According to the results obtained, the algorithm proposed here shows 10 times better performance when compared to the traditional region-growing method.