Indoor-Outdoor 3D Reconstruction Alignment
Structure-from-Motion can achieve accurate reconstructions of urban scenes. However, reconstructing the inside and the outside of a building into a single model is very challenging due to the lack of visual overlap and the change of lighting conditions be
- PDF / 3,063,957 Bytes
- 16 Pages / 439.37 x 666.142 pts Page_size
- 41 Downloads / 256 Views
ETH Z¨ urich, Z¨ urich, Switzerland {acohen,jsch}@inf.ethz.ch UNC Chapel Hill, Chapel Hill, USA 3 Microsoft, Redmond, USA
Abstract. Structure-from-Motion can achieve accurate reconstructions of urban scenes. However, reconstructing the inside and the outside of a building into a single model is very challenging due to the lack of visual overlap and the change of lighting conditions between the two scenes. We propose a solution to align disconnected indoor and outdoor models of the same building into a single 3D model. Our approach leverages semantic information, specifically window detections, in multiple scenes to obtain candidate matches from which an alignment hypothesis can be computed. To determine the best alignment, we propose a novel cost function that takes both the number of window matches and the intersection of the aligned models into account. We evaluate our solution on multiple challenging datasets.
1
Introduction
Recent progress in the area of 3D reconstruction enables the generation of largescale [12] and detailed outdoor models [25], as well as accurate indoor reconstructions [9] and their floor-plans [2,19]. The resulting 3D models are useful for a wide range of applications, from virtual tourism [16,27], visualization of apartments for real estate [19], cultural heritage [7,23,29], and image-based localization [18,30], to real-time camera pose tracking on mobile devices for Augmented Reality [22], and autonomous navigation [20]. Ideally, a single joint reconstruction of the interior and exterior is desirable as it would, for example, enable a user to seamlessly enter buildings in a virtual city model rather than only exploring the outside. Similarly, a combined model would allow autonomous robots to easily transition between the indoor and outdoor world. However, state-of-the-art approaches often fail to reconstruct both parts into a single 3D model. Obtaining a joint indoor-outdoor model is hard for multiple reasons: on the one hand, the indoor and outdoor parts of a scene typically exhibit a weak connection through a limited number of visual observations such as doorways or Electronic supplementary material The online version of this chapter (doi:10. 1007/978-3-319-46487-9 18) contains supplementary material, which is available to authorized users. c Springer International Publishing AG 2016 B. Leibe et al. (Eds.): ECCV 2016, Part III, LNCS 9907, pp. 285–300, 2016. DOI: 10.1007/978-3-319-46487-9 18
286
A. Cohen et al.
Fig. 1. The proposed method aligns disconnected Structure-from-Motion reconstructions of the inside and outside to produce a single 3D model of a building. Our approach also handles incomplete reconstructions and multiple indoor models (c.f. right model)
windows. As a result, great care must be taken when capturing data to ensure enough visual overlap for feature matching and to prevent the models from being disconnected [28]. This problem is often aggravated by the fact that there can be a strong change in illumination in transition areas. In practice, Structurefrom-Motio
Data Loading...
