Abstract
Bundle adjustment plays an significant role in SLAM (simultaneous localization and mapping), which is utilized in both front-end visual odometry and global back-end optimization. In many SLAM systems, bundle adjustment is employed to estimate the location of 3D landmarks and 6 DOF camera pose. However, as the dimension of the optimization variable increases, bundle adjustment consumes more and more time in SLAM, mainly in the iterative process. In this paper, an improved algorithm for this large scale bundle adjustment problem has been proposed. Firstly, according to the pose consensus, the classic algorithm ADMM (alternating direction method of multipliers) is introduced into the bundle adjustment problem. Secondly, for the non-convex optimization problem, the sub problem optimization method is introduced, and convergence and stopping criteria of the algorithm are discussed. Finally, the semi-dense direct method visual odometry for verification is implemented, and the experiments prove that the improved bundle adjustment algorithm has a speed advantage and can be applied to BAL (bundle adjustment in the large) problem.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call