Abstract

This paper describes an automatic pipeline that is able to take a set of unordered range images and align them into a full 3D model. A global voting scheme is employed for view matching, inspired by 2D techniques for image mosaicing. Then a multiple view registration approach is introduced, which aims at optimizing the alignment error simultaneously for all the views. Experiments demonstrate the effectiveness of the method.

ABSTRACT:A novel multi-view stereo reconstruction method is presented. The algorithm is focused on accuracy and it is highly engineered with some parts taking advantage of the graphics processing unit. In addition, it is seamlessly integrated with the output of a structure and motion pipeline. In the first part of the algorithm a depth map is extracted independently for each image. The final depth map is generated from the depth hypothesis using a Markov random field optimization technique over the image grid. An octree data structure accumulates the votes coming from each depth map. A novel procedure to remove rogue points is proposed that takes into account the visibility information and the matching score of each point. Finally a texture map is built by wisely making use of both the visibility and the view angle informations. Several results show the effectiveness of the algorithm under different working scenarios.

No abstract

No abstract