Three-dimensional computer vision has become an important tool in industrial inspection, an area where it has been applied most successfully. Mainly two-dimensional inspection problems have been solved to date. Three-dimensional geometric inspection to typical engineering tolerances is just becoming possible with computer vision techniques, and a main objective of this book is to show how this can be achieved.

An important subtask in inspection is object recognition, and a substantial part of this book is devoted to a consideration of various current object recognition strategies. The volume is of great value for a wide range of readers because it provides exhaustive coverage of most problems arising in computer vision, models, and inspection. This book both surveys existing methods and presents new results in these fields. Every relevant chapter has a set of exercises that makes the book useful as a classroom text.

The authors underline and analyze the problems met in their practice, both with acquisition systems and with the design of the data structures and algorithms conceived for different goals. The first chapter presents the scope of the book and the visual inspection system discussed, assembled at the British Aerospace Sowerby Research Center.

Chapter 2 describes the first stage of any vision system, image acquisition. This chapter addresses many aspects of capturing an image in a form suitable for analysis by computer. Both two- and three-dimensional images are discussed.

The third chapter is concerned with image processing methods consisting primarily of the manipulation of data contained within an image to improve or enhance certain properties of the image. Finally, the authors discuss image compression, because of its importance in data storage and the transmission of data over computer networks.

Chapter 4 discusses the types of high-level information that may be extracted from images and the methods employed to extract and represent such features. A lengthy description of edge detection methods, an important part of image processing, is included. The fifth chapter shows how various features can be grouped to form higher-level feature descriptions, as intermediate steps along a path ending with the recognition of objects present in the scene.

Chapter 6 is an introduction to Part 2 of the book, which deals with three-dimensional object recognition. Chapter7 discusses various methods of representing the geometric information about objects needed in geometric modeling. In conclusion, the authors deduce that for most computer vision applications, boundary representation is the most promising method of modeling solid objects.

Chapter 8 describes a variety of methods to solve the problem of matching a scene description to a model description; each method is suitable for only a few tasks that may require a recognition system. In the following sections, brief algorithmic details of some popular matching strategies are given.

In chapter 9, many of the concepts discussed in the preceding chapter are used to develop a model-based matching algorithm. The emphasis is on accuracy to satisfy the requirements of inspection tasks. The approach is based on the Faugeras and Hebert matching technique. By using artificial data, the authors have shown the expected errors from this approach, which are sufficiently small to make geometric inspection feasible.

Chapter 10 is an introduction to Part 3, which brings together the techniques required for the ultimate goal of an automated vision-based three-dimensional inspection system. This part shows how the matching stage can be used as part of an overall system for performing geometric inspection of the observed object. Inspection involves comparing the object with some object model that describes the relevant features of the object. The particular inspection task described uses a geometric model of the object together with a suitable description of the tolerance information. The discussion includes the choice of sensing equipment and segmentation and matching algorithms.

Chapter 11 deals with the problems of actually measuring and testing features for conformity to tolerances. Chapter 12 discusses inspection of the features visible in a single scene. An overview of geometric inspection strategies is given. Then the authors describe a method for inspecting a single view of an object to decide if the features are within geometric tolerances.

The ideas discussed in chapter 13 aim to demonstrate the feasibility of inspection using multiple scenes. Single-axis inspection of objects to typical inspection tolerances has been demonstrated. The strategies outlined here are rudimentary, although they have been shown to work for a small class of objects.

Chapter 14 is a brief discussion of how computer vision techniques can be applied to other inspection tasks in areas where visual inspection systems can be utilized. Applications of visual inspection in printed circuit boards and the food and agricultural industries are described.

Chapter 15 presents conclusions and discusses future work. The book demonstrates that geometric inspection to standard engineering tolerances is feasible using a three-dimensional vision system employing a range of computer vision techniques. The references are adequate, complete, and topical.