Towards Real-Time Trinocular Visual Odometry

Abstract:

Pose estimation of multi-camera rig which has not enough overlapping field of views for the stereo, is generally computationally expensive due to the offset of camera center and the bundle adjustment algorithm. We proposed a divide and conquer approach, which reduces the trinocular visual odometry problem to five monocular visual odometry problems, one for each individual camera sequence and two more using features matched temporally from consecutive images from the center to the left and right cameras, respectively. While this approach provides high accuracy over long distances in outdoor environment without requiring any additional sensors, it is computationally expensive, preventing real-time operation. In this paper, we evaluate trading off image resolution and frame rate to speed up computation with accuracy. Results show that scaling images down to a quarter of Full HD resolution can speed up computation by two orders of magnitude, while still providing acceptable accuracy, whereas dropping frames quickly deteriorates performance.

• Periodical: Applied Mechanics and Materials (Volumes 490-491)
• Pages: 1424-1429
• DOI: https://doi.org/10.4028/www.scientific.net/AMM.490-491.1424
• Online since: January 2014
• Authors: Jae Heon Jeong, Nikolaus Correll
• Keywords: Fundamental Matrix, Multi-Camera, RANSAC, Visual Odometry

References

[1] J. Jeong, J. Mulligan and N. Correll, Trinocular Visual Odometry with Minimal Overlap, in Proc. WORV (2013).

DOI: 10.1109/worv.2013.6521943

[2] T. Oskiper, Z. Zhu, S. Samarasekera and R. Kumar, Visual Odometry System using Multiple Stereo Cameras and Inertial Measurement Unit, in Proc. CVPR'07 (2007).

DOI: 10.1109/cvpr.2007.383087

[3] B. Triggs, P. Mclauchlan, R. Hartley and A. Fitzgibbon, Bundle Adjustment–a Modern Synthesis, in ICCV '99 (1999).

[4] K. Konolige and W. Garage, Sparse Bundle Adjustment, in BMVC (2010), p.1–11.

[5] J. Kim, R. Hartley, J. Frahm and M. Pollefeys, Visual Odometry for Non-overlapping Views using Second-order Cone Programming, in Proc. ACCV07 (2007), p.353–362.

DOI: 10.1007/978-3-540-76390-1_35

[6] J. H. Kim, H. Li and R. Hartley, Motion Estimation for Multi-camera Systems using Global Optimization, in CVPR'08. (2008), p.1 –8.

[7] J. Kim, H. Li and R. Hartley, "Motion Estimation for Nonoverlapping Multicamera Rigs: Linear Algebraic and l1 Geometric Solutions, in PAMI (2010) vol. 32, no. 6, p.1044 – 1059.

DOI: 10.1109/tpami.2009.82

[8] T. Kazik, L. Kneip, J. Nikolic, M. Pollefeys and R. Siegwart, Realtime 6d Stereo Visual Odometry with Non-overlapping Fields of View, in CVPR'12 (2012), p.1529 –1536.

DOI: 10.1109/cvpr.2012.6247843

[9] M. Ragab and K. Wong, Multiple Nonoverlapping Camera Pose Estimation, in ICIP (2010) p.3253 –3256.

[10] Y. Dai, M. He, H. Li and R. Hartley, Factorization-based Structureand-motion Computation for Generalized Camera Model, in ICSPCC(2011) p.1 –6.

DOI: 10.1109/icspcc.2011.6061822

[11] J. S. Kim, M. Hwangbo and T. Kanade, Spherical Approximation for Multiple Cameras in Motion Estimation: Its Applicability and Advantages, Computer Vision and Image Understanding, vol. 114, no. 10, p.1068 – 1083, (2010).

DOI: 10.1016/j.cviu.2010.07.005

[12] M. Kaess and F. Dellaert, Probabilistic Structure Matching for Visual Slam with a Multi-camera rig, Comput. Vis. Image Underst., vol. 114, no. 2, p.286–296 (2010).

DOI: 10.1016/j.cviu.2009.07.006

[13] G. Carrera, A. Angeli and A. J. Davison, Lightweight Slam and Navigation with a Multi-camera rig, (2011).

[14] D. Nist´er, O. Naroditsky and J. Bergen, Visual Odometry for Ground Vehicle Applications, Journal of Field Robotics, vol. 23, (2006).

DOI: 10.1002/rob.20103

[15] M. A. Lourakis and A. Argyros, SBA: A Software Package for Generic Sparse Bundle Adjustment, ACM Trans. Math. Software, vol. 36, no. 1, p.1–30 (2009).

DOI: 10.1145/1486525.1486527