Large-scale 3D Scene Reconstruction Research Articles

Efficient convex optimization-based texture mapping for large-scale 3D scene reconstruction

Texture mapping is a key step in large-scale 3D scene reconstruction, which can greatly enhance visual reality of the reconstructed scenes. However, existing techniques are unable to accomplish this task efficiently due to the high computational complexity of reconstructing large-scale real-world 3D scenes. In this work, we propose a new efficient convex optimization-based approach, i.e. the mesh-based continuous max-flow method, which can be easily implemented and accelerated upon a modern parallel computing platform, e.g. GPU. Particularly, an Markov Random Fields (MRF) based model, i.e. Potts model, is introduced to mathematically formulate the key specific view selection problem; we show that the challenging combinatorial optimization problem can be efficiently solved by resolving its convex relaxation, which recovers textures from images with the proposed duality-based continuous max-flow approach. In addition, visual effects of sharpness and deformation are utilized to define a criterion of evaluating texture quality effectively, and a large 3D triangular mesh is partitioned into structural components so as to reduce memory consumption of the proposed algorithm. The proposed mesh-based continuous max-flow approach for large-scale texture mapping demonstrates its outperformance over state-of-the-art methods, over large-scale public datasets, in both numerical efficiency and visual quality; meanwhile, our GPU-accelerated algorithm can yield 3D textured models with high quality from complex large-scale scenes in minutes.

Efficient Large-Scale Structure From Motion by Fusing Auxiliary Imaging Information.

One of the potentially effective means for large-scale 3D scene reconstruction is to reconstruct the scene in a global manner, rather than incrementally, by fully exploiting available auxiliary information on the imaging condition, such as camera location by Global Positioning System (GPS), orientation by inertial measurement unit (or compass), focal length from EXIF, and so on. However, such auxiliary information, though informative and valuable, is usually too noisy to be directly usable. In this paper, we present an approach by taking advantage of such noisy auxiliary information to improve structure from motion solving. More specifically, we introduce two effective iterative global optimization algorithms initiated with such noisy auxiliary information. One is a robust rotation averaging algorithm to deal with contaminated epipolar graph, the other is a robust scene reconstruction algorithm to deal with noisy GPS data for camera centers initialization. We found that by exclusively focusing on the estimated inliers at the current iteration, the optimization process initialized by such noisy auxiliary information could converge well and efficiently. Our proposed method is evaluated on real images captured by unmanned aerial vehicle, StreetView car, and conventional digital cameras. Extensive experimental results show that our method performs similarly or better than many of the state-of-art reconstruction approaches, in terms of reconstruction accuracy and completeness, but is more efficient and scalable for large-scale image data sets.