An improved bundle adjustment model and algorithm with novel block matrix partition method

Abstract

Sparse bundle adjustment is widely applied in computer vision and photogrammetry. However, existing implementation is based on the model of <i>n</i> 3D points projecting onto <i>m</i> different camera imaging planes at <i>m</i> positions, which can't be applied to commonly monocular, binocular or trinocular imaging systems. A novel design and implementation of bundle adjustment algorithm is proposed in this paper, which is based on <i>n</i> 3D points projecting onto the same camera imaging plane at <i>m</i> positions .To improve the performance of the algorithm, a novel sparse block matrix partition method is proposed. Experiments show that the improved bundle adjustment is effective, robust and has a better tolerance to pixel coordinates error.