Vocal 2018 [antikvár]

3D Reconstruction with Depth Prior Using Graph-Cut * Hichem Abdellali and Zoltan Kato Institute of Informatics, University of Szeged, H-6701 Szeged, PC. BOX 652., Hungary Email: {hichem, kato}@inf.u-szeged.hu Abstract. In this paper we propose a novel graph-cut based 3D reconstruction method which is able to take into account partially available depth data as a prior. We explored the possibiUty of using a prior information to achieve an efficient 3D scene reconstruction using MRF Modelling and graph-cut, which represent the disparity as an energy function. We formulate the energy in two representations: 1) assignment-based, which yields a standard binary energy; as well as 2) a multi-label one which yields a non-binary energy. Both representations have its advantages and disadvantages, which are analysed in detail through various experiments on the Middlebury stereo data set. Results show, that the use of depth prior information from different sources produces better 3D reconstructions. Keywords: 3D Reconstruction • Graph-Cut • MRF Modelling. 1 Introduction By using a pair of rectified binocular images, it is possible to reconstruct the 3D scene by finding dense correspondences between the images and building a disparity map. Depth information is useful for many application like modeling, monitoring, urban mapping, and autonomous navigation. Nowadays, various depth sensors are available to capture a 3D scene, like Time-of-flight devices, or Lidar. however, these are sensitive to lighting conditions and require a special setup, while stereo camera systems are more fiexible, cheaper and suitable for disparity estimation. In this paper, we propose a new graph representable energy function based on the previous work of [4,5,3], with a new additional term which takes into account a prior disparity map collected from other sources. Introducing this depth prior provides a soft way to improve the disparity. Recently, * This work was partially supported by the NKFI-6 fund through project K120366; "Integrated program for training new generation of scientists in the fields of computer science", EFOP-3.6.3-VEKOP-16-2017-0002; the Research & Development Operational Programme for the project "Modernization and Improvement of Technical Infrastructure for Research and Development of J. Selye University in the Fields of Nanotechnology and Intelligent Space", ITMS 26210120042, co-funded by the European Regional Development Fund.