Building 3D Object Reconstruction System Based on Color-Depth Image Sequences

In this paper, we present a 3D object reconstruction system based on color and depth image sequences. The main intention is to build a sufficient system that can digitize the shape and color texture of real objects. Our contributions are the combination of individual processes into one unified procedure and some improvements on these methods. The system consists of four main phases. First, we improve the separation of object from background by performing a saliency map on depth image. Next, we represent the surface of object by a global point cloud and use Poisson reconstruction method to reconstruct the surface mesh. Finally, we propose neighborhood interpolation mapping technique for assigning color of reconstructed model. The experiments conducted on practical datasets have shown that our system is able to scan, reconstruct and simulate real objects completely.