Geometric Computation for Machine Vision

Geometric Computation for Machine Vision Synopsis

Machine vision is the study of how to build intelligent machines which can understand the environment by vision. Among many existing books on this subject, this book is unique in that the entire volume is devoted to computational problems, which most books do not deal with. One of the main subjects of this book is the mathematics underlying all vision problems - projective geometry, in particular. Since projective geometry has been developed by mathematicians without any regard to machine vision applications, our first attempt is to `tune' it into the form applicable to machine vision problems. The resulting formulation is termed computational projective geometry and applied to 3-D shape analysis, camera calibration, road scene analysis, 3-D motion analysis, optical flow analysis, and conic image analysis. A salient characteristic of machine vision problems is that data are not necessarily accurate. Hence, computational procedures defined by using exact relationships may break down if blindly applied to inaccurate data. In this book, special emphasis is put on robustness, which means that the computed result is not only exact when the data are accurate but also is expected to give a good approximation in the prescence of noise. The analysis of how the computation is affected by the inaccuracy of the data is also crucial. Statistical analysis of computations based on image data is also one of the main subjects of this book.