Bundle ICP with Virtual Depth for Hand-Held 3d Scanner

RGB-D cameras provide both color information and per-pixel depth information. The richness of their data present an attractive opportunity for mobile application such as Augmented Reality (AR) and 3D scanner. In this paper, we propose a general-purpose hand-held 3D scan system that combines a iterative closest point (ICP) algorithm based on a large amount of virtual information for accuracy with the advantage of a graph-based reconstruction system for robustness. First, the image graph is built with correlation between the images and the camera motion is estimated with proposed Bundle ICP method by fusing the visual information and virtual depth information. After registering the image incrementally, a sparse reconstruction model is created with global pose consistence. Finally, all the depth data from registered image are fused into a single global volume to reconstruct surface. The results of the experiment show that the proposed method has better performance than the traditional vision-based structure from motion (SFM) and tracking based reconstruction methods.