![]() ![]() ![]() #Wise memory optimizer heise manual#The final reconstructions provide fine details, but contrast to our system, these approaches involve manual this system limits the range of the monocular camera to intervention and editing at different stages of the pipeline near-field reconstructions. support model capture using stationary lidar and motion with 2D-regularized depth estimates from sequential image photogrammetry (Moussa et al. R2: Range R3: Data R4: Sensors 2 2 DTAM 1m 1 m 100+ img/m Mono Kinect Fusion 7 m 2 m ––––||–––– RGB-D Kintinuous ‘ 3 m ––––||–––– RGB-D Voxel hashing ‘ 4 m ––––||–––– RGB-D 2 2 BOR G ‘ 50 m 0.05 img/m Mono, stereo or lidar 2D regularization. ![]() System capabilities comparison Technique R1: Env. Together these strands, Email: 2 638 The International Journal of Robotics Research 41(6) The International Journal of Robotics Research 00(0) Table 1. 2012), upon which the most advanced Stefan Sa˘ ftescu, Oxford Robotics Institute, University of Oxford, 17 Parks algorithms rely, have become accessible for robotics and Road, Oxford OX1 3PJ, UK. The Oxford Robotics Institute, University of Oxford, UK RACE, UK Atomic Energy Authority, Abingdon, UK ever-strengthening and broadening theoretical foundations of continuous optimization (Chambolle and Pock 2011 Corresponding author: Goldluecke et al. (2010) reconstructed sections of urban scenes from unstructured photo collections. Earlier, large-scale efforts tectural and archeological sites with methods designed to such as those of Pollefeys et al. Over the past few years, the development of 3D recon- Several cultural heritage projects fuse multimodality struction systems has undergone an explosion facilitated by sensor data to build high-fidelity representations of archi- the advances in GPU hardware. 2015) with an eye on small-scale reconstruction. The compu- motivated by the recent mobile phone and tablet develop- tational and memory requirements of large dense models ment (Klingensmith et al. The most general approaches are leads to better robots, but this comes at a cost. Better cal use in mapping applications such as autonomous maps make for better understanding better understanding driving or inspection. This article is about the efficient generation of dense, However, the state of the art of many dense 3D recon- colored models of very-large-scale environments from struction systems rarely considers scalability for the practi- stereo cameras, laser data, or a combination thereof. Introduction and previous work large-scale 3D dense reconstructions as shown in Figure 2. Keywords Dense reconstruction, regularization, mapping hardware and theory, allow us to build systems that create 1. We demonstrate our system in practice by reconstructing the inside of the EUROfusion Joint European Torus (JET) fusion reactor, located at the Culham Centre for Fusion Energy (UK Atomic Energy Authority) in Oxfordshire. We provide statistics for the metric errors in all surfaces created compared with those measured with 3D lidar as ground truth. These latter datasets see us operating at a combined scale and accuracy not seen in the literature. We evaluate our system using the Stanford Burghers of Calais, Imperial College ICL-NUIM, Oxford Broad Street (released with this paper), and the KITTI datasets. We take the unusual step to apply a learned correction mechanism that takes the global context of the reconstruc- tion and adjusts the constructed mesh, addressing errors that are pathological to the first-pass camera-derived reconstruction. Our pipeline does not end with regulari- zation. Our regularizer reduces the median error between 27% and 36% in 7.3 km of dense reconstructions with a median accuracy between 4 and 8 cm. In addition, because of the paucity of surface observations by the camera and lidar sensors, we regularize over both two (camera depth maps) and three dimensions (voxel grid) to provide a local contextual prior for the reconstruction. We use a compressed 3D data structure, which allows us to operate over a large scale. Our BOR G system fuses data from multiple sensor modalities (cameras, lidars, or both) and regularizes the resulting 3D model. ![]() We provide the theory and the system needed to create large-scale dense reconstructions for mobile-robotics applications: this stands in contrast to the object-centric reconstructions dominant in the literature. Tanner, Michael Piniés, Pedro Paz, Lina María Săftescu, Ştefan Bewley, Alex Jonasson, Emil Newman, Paul Large-scale outdoor scene reconstruction and correction with vision Large-scale outdoor scene reconstruction and correction with vision ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |