Optimizing camera positions for multi-view 3D reconstruction

In the last decades, a number of high-quality multi-view 3D reconstruction algorithms have been developed to reconstruct a 3D object model with a collection of images captured from different points of view. Intuitively, with more images from different viewpoints, more information can be utilized for the reconstruction algorithms. However, the acquisition effort, storage and computing cost will grow correspondingly with the number of the images. Furthermore, the possible noise and unrelated information introduced by additional images make the reconstruction more challenging. In this work, the relation between the reconstruction quality and the distribution of camera positions is analyzed and a new camera positioning strategy based on the properties of the synthetic object surface is proposed to achieve the same or even better reconstruction quality with equal or fewer viewpoints.