This collection of methods and techniques concerns the so-called structure from motion (SfM) problem--that is, the 3D reconstruction of geometries and camera positions/orientations, starting from a sequence of 2D images in room-sized environments. Over the past several decades, many research efforts have focused on this problem, also referred to as monocular simultaneous localization and mapping (SLAM), and this book addresses the SfM problem as an unsupervised 3D sparse points reconstruction, in particular using the extended Kalman filter.

The chapters present a detailed description of “a robust [framework] for sequential scene and camera motion estimation,” able to ingest image sequences at 30 frames per second (fps) in real time. Topics of the book range from “a projective point model [to] an efficient ... search for correspondences and algorithms for model selection and internal self-calibration.”

The content is derived from several peer-reviewed conferences and journals, and a final outcome of this research has been Civera’s (one of the authors) PhD thesis. While the book is written, well motivated, and proven from a pure research point of view, there is a lack of real practical cases where the framework has been applied. This in some ways casts doubt on the actual usability of the methods.

The lack of tests in real-world scenarios limits the importance of this book. However, in my opinion, it is still a good read for researchers and PhD students in computer vision and robotics areas, because it provides an interesting point of view on how to attack and solve the SfM problem.