A Few Steps Towards 3D Active Vision

Citation

Vieville, T. (1998), "A Few Steps Towards 3D Active Vision", Sensor Review, Vol. 18 No. 2, pp. 143-143. https://doi.org/10.1108/sr.1998.18.2.143.1

Number 33 in the Springer series in information sciences, this modestly titled volume provides a grounding in the methods of 3D active vision both in itself, and for robotic applications in particular.

3D active vision is the method by which visual information is used as feedback into the control of the visual sensor’s physical location, its optical components and processing, best illustrated by considering the text’s example of a robotic “head” where a stereoscopic vision system is mounted on a movable platform or “neck”.

The book is structured much like a thesis, setting out the 3D active vision problem by means of a survey of relevant work and literature, before going on and offering specific, but also hardware‐independent solutions in the remainder of the text.

The case for developing the technology is presented in relation to the use of more conventional 2D and static systems, and the principal requirements are set out for a viable implementation.

The design of a working system (based on the robotic head) is discussed in detail, with experimental data and some photographs of the equipment to illustrate the work on the project.

The theme of automatic calibration of the equipment is taken up in more detail, the development of the theory culminating in the presentation of a suitable algorithm and experimental data to validate the technique.

An unusual departure from the norm is the inclusion of additional non‐visual sensors to the system, where the work reports the inclusion of inertial sensors on the moving parts of the system to add additional clues to orientation, movement and the 3D perception of the environment.

The final topic before the conclusion to the work presents a methodology for the geometric interpretation of structure and motion from monocular image sequences. A complete set of references is included, organised by chapter.

This is obviously a relatively specialised text aimed primarily at the researcher, though the author’s approach is to develop the theory from a reasonably basic viewpoint before describing an implementation and discussing relevant experimental results.

A grasp of the mathematics underlying the like of camera calibrations and inverse kinematics would be necessary to dissect the entirety of the theory within the work; however, this is not a prerequisite for reading the majority of the book and picking up on a number of original concepts.