Object recognition is currently an area of considerable research interest. This book touches on most aspects of the recognition problem, with the primary goal of considering components of the recognition problem while describing a detailed exploration of one aspect of object recognition. The following question is specifically investigated: What is the role of geometric measurements and constraints in object recognition and localization? The author is interested in understanding how the shapes of objects can be used to determine which objects from a library of possible objects are actually present in a scene, to determine the correspondence between data features and object features, and to determine the position of an object in the scene. The geometric measurements are intended to capture aspects of an object’s shape as perceived by a sensor and any changes in that perceived shape as the object is transformed in the scene. Grimson is particularly interested in measurements that are invariant under the set of allowed transformations, as these measurements will sharply constrain the solutions to all three subproblems: the set of possible object models in the scene, the set of possible correspondences between scene features and model features, and the set of possible positions of an object in the scene.

The book is mainly written as a monograph, but also includes the work of other researchers in the field (including T. Lozano-Perez and D. P. Huttenlocher, who also wrote some parts of the book). An extensive list of about 300 references is given at the end of the book.

After a concise introduction, chapter 2 explores the role of geometric constraints and alternative methods of searching for instances of an object in the data. Chapter 3 concentrates on the constrained search approach. This chapter lays out a general framework for constrained search. In chapters 4 and 5, details of different types of geometric constraints that can be plugged into that framework are provided. Chapter 6 considers the problem of actually finding the position of an object associated with an interpretation and verifying that the interpretation is globally consistent.

One of the main problems in constrained search approaches to recognition is controlling the inherent combinatorial explosion associated with the search. Variations of the search method used to find interpretations, and their effect on the combinatorics, are considered in chapter 7. Chapter 8 considers other methods for controlling this explosion, mainly by restricting the portions of the search space to be explored. Empirical data summarizing the effects of these choices are given in chapter 9.

The second part of the book gives a formal way of examining the various methods. In particular, chapters 10 through 13 develop a formal model of the recognition method and derive analytic results on the complexity of constrained search approaches to recognition. These results carry some implications concerning the relative difficulty of different parts of the recognition problem, which are discussed in chapter 14.

The final part of the book deals with various extensions of the basic methods developed in the first part and analyzed in the second. Chapter 15 deals briefly with the problem of recognition from libraries of objects, chapter 16 discusses extensions from rigid objects to broader classes of objects, chapter 17 briefly discusses the role of grouping in recognition, and chapter 18 explores the idea of sensing strategies. Finally, chapter 19 briefly describes some representative applications of these recognition methods.

This book will be useful to the wide audience of theoretical and practical researchers in object recognition, and to graduate students specializing in artificial intelligence. It is well written and easy to read, and contains many illustrations and examples. I recommend it as a textbook in the area.