3D surface reconstruction based on Kinect

In this paper, a simple and efficient 3D surface reconstruction algorithm which can be implemented on mid or low-end computers is proposed. In this algorithm, the 3D images are obtained by a depth motion sensing device named Kinect. Data from region of interests are obtained by segmenting images whose k-neighbor relationship is established with a k-dimensional tree (KD-tree). After patching the point cloud holes with a bicubic spline function and a triangular mesh, the 3D surface of the object is reconstructed. Furthermore, the neighborhoods defined differently are used to connect regions since several non-connected regions exist in the point cloud. Experimental results show that the proposed algorithm based on the Kinect platform can be used to reconstruct complete and smooth surface even if the point cloud is non-connected in space and some of the captured data are lost.