Lourakis and Argyros address the common problem of bundle adjustment that is traditionally associated with--and is an imperative component of--three-dimensional (3D) scene reconstruction. In this context, bundle adjustment can be described as a minimization problem, where the 3D structure and viewing (that is, extrinsic and sometimes intrinsic) parameters are optimized, using a set of correspondences between a group of two-dimensional (2D) images and the 3D geometry of the scene. The optimal values are found by “minimizing the reprojection error between the observed and predicted image points,” using a nonlinear least-squares algorithm [1].

Bundle adjustment is a well-known problem, in particular to the computer vision community, and is already a well-studied subject. However, this paper presents novel ideas on how to improve the performance and efficiency of the optimization by exploiting the fact that “the normal equations matrix has a sparse block structure owing to the lack of interaction among parameters for different 3D points and cameras” [1]. The paper proposes solutions to the important problem of reducing computational complexity, at a time when 3D reconstructions are becoming increasingly demanding in terms of dimensionality.

The paper is well written and its organization makes it very easy to follow. It begins with a thorough description of the problem and a discussion of the existing solutions and their limitations. Next, the authors present their strategy in great detail, using simple examples that can be easily generalized to higher dimensions. The interested reader can then follow through with the implementation details of the software package SBA; this is finally validated experimentally to show that the computational performance is indeed greatly improved.