Structure from Motion and 3D Reconstruction
Structure from motion is a fundamental technical that forms one of the corner stones of the vision-based 3D reconstruction, localisation and navigation systems. The robustness and efficiency of the structure from motion algorithm is crucial for the reliably and performance of such systems. Most state-of-the-art algorithms rely on repetitively apply expensive bundle adjustment to incrementally align the camera motion and scene structure. Systems based on such algorithms often fall short of the efficiency and low power requirement of many localisation and navigation applications.
In our current project, we study the fundamental stability of different parameters in structure from motion, and explore different ways to reformulate the problem, such that the computational resources will be allocated adaptively according to the difficulty of reliably estimating these parameters. With these exploration, we hope to gain a deeper understanding of the bottle neck of modern structure from motion algorithms in different application scenarios.
- W.-Y. Lin, S. Liu, N. Jiang, M. N. Do, P. Tan, and J. Lu, “RepMatch: Robust Feature Matching and Pose for Reconstructing Modern Cities,” European Conf. Computer Vision (ECCV), Amsterdam, Oct. 2016. [pdf][supplementary]
- N.-J. Jiang, W.-Y. Lin, M. N. Do, and J. Lu, “Direct Structure Estimation for 3D Reconstruction,” IEEE Int. Conf. Computer Vision and Pattern Recognition (CVPR), Boston, MA, Jun. 2015. [pdf][code]
- N.-J. Jiang, Z.-P. Cui, P. Tan, “A Global Linear Method for Camera Pose Registration”, IEEE International Conference on Computer Vision (ICCV), 2013. (Oral)
- Our related work:
- W.-Y. Lin, M. Cheng, J. Lu, H. Yang, M. N. Do, and P. H. S. Torr, “Bilateral Functions for Global Motion Modeling,” in Proc. Eur. Conf. Computer Vision (ECCV), Zurich, Switzerland, Sep. 2014. [pdf]
- W.-Y. Lin, F. Wang, M. Cheng, S.-K.Yeung, P. Torr, M. N. Do, and J. Lu, “CODE: Coherence Based Decision Boundaries for Feature Correspondence,” IEEE Trans. on Pattern Analysis and Machine Intelligence (TPAMI), 2017. [preprint]