Fast Robust Large-scale Mapping from Video and Internet Photo ...

More documents

Recommendations

Info

Figure 4: Left: view of the 3D model and the 9025 registered cameras. Right: model from the interior of Ellis Island erroneously labeled as “Statue of Liberty”. extending the registration to all cluster images in the clusters corresponding to the registered iconics. This process takes advantage of feature matches between the non-iconic images and their respective iconics. This process again uses the 2D matches to the iconic to determine the 2D-3D correspondences for the image. It then applies ARRSAC (Section 4.1.2 to determine the camera s registration. Given the incremental image registration during this process drift inherently accumulates. Hence we periodically apply bundle adjustment to perform a global error correction. In our current system bundle adjustment is performed after 25 images have been added to a sub-model. This saves significant computational resources while in practice keeping the accumulated error at a reasonable level. Detailed results of our 3D reconstruction algorithm are shown in Figures 4-5. 5. Dense Geometry The above described methods for video and for photo collections provide a registration for the camera poses of all video frames or photographs. These external and internal camera calibrations are the output of the sparse 28
Figure 5: Left: 3D reconstruction of the San Marco Dataset with 10338 cameras. Right: Reconstruction from the Notre Dame dataset with 1300 registered cameras. Dataset Gist Gist clustering Geometric Re-clustering registered views total time Liberty 4 min 21 min 56 min 3.3 hr 13888 4 hr 42 min SM 4 min 19 min 24 min 2.76 hr 12253 3 hr 34 min ND 1 min 3 min 22 min 1.0 hr 3058 1 hr 28 min Table 1: Computation times for the photo collection reconstruction for the Statue of Liberty dataset, the San Marco dataset (SM) and the Notre Dame dataset (ND). reconstruction step of both systems. They allow for the photo collections to generate an image based browsing experience as shown by the “Photo- Tourism” paper [51, 52]. The camera registration can also be used as input to dense stereo methods. The method of [83] can extract dense scene geometry but with significant computational effort and therefore only for small scenes. Next we will introduce our real-time stereo system for video streams capable of reconstructing large scenes. We adopt a stereo/fusion approach for dense geometry reconstruction. 29
Page 1 and 2: Fast Robust Large-scale Mapping fro
Page 3 and 4: Figure 1: The left shows an overvie
Page 5 and 6: use statistical models to combine d
Page 7 and 8: a probabilistic way and the final s
Page 9 and 10: streams of multiple cameras mounted
Page 11 and 12: In the case of available GPS data o
Page 13 and 14: 4.1. Camera Pose from Video Our sys
Page 15 and 16: complexity. Our recently proposed A
Page 17 and 18: surements we must normalize all of
Page 19 and 20: VIP-features [68]. To avoid a compu
Page 21 and 22: an active research topic in the com
Page 23 and 24: achieve high computational performa
Page 25 and 26: for each image in the cluster to re
Page 27: can be added to the 3D model. This
Page 31 and 32: from 11 images (10 matching, 1 refe
Page 33 and 34: Original Mesh Simplified Mesh Textu
Page 35 and 36: 7. Conclusions In this paper we pre
Page 37 and 38: [12] T. Berg, D. Forsyth, Animals o
Page 39 and 40: [30] Y. Jing, S. Baluja, H. Rowley,
Page 41 and 42: objects from multiple range images,
Page 43 and 44: [63] M. Fischler, R. Bolles, Random
Page 45 and 46: Marquardt algorithm, Tech. Rep. 340

Fast Robust Large-scale Mapping from Video and Internet Photo ...

Create successful ePaper yourself

Delete template?

Save as template?