A Unified Framework for Quasi-Linear Projective, Affine and Metric Bundle Adjustment and Pose Estimation

Obtaining 3d models from large image sequences is a major issue in computer vision. One a the main tools used to obtain accurate structure and motion estimates is bundle adjustment. Bundle adjustment is usually performed using non-linear Newton-type optimizers such as LevenbergMarquardt which might be quite slow when handling a large number of points or views. We propose an algorithm for bundle adjustment based on quasi-linear optimization. The method is straightforward to implement and relies on solving weighted linear systems obtained as simple functions of the input data. The algorithm acts in projective or metric space, using full perspective or affine projection. It can handle any number of views and points as well as missing data. We also provide a quasi-linear n ≥ 3-points pose estimator. Experimental results on simulated and real data show that the algorithm is as accurate as standard techniques while requiring less computational time to converge.