Embedded computer vision is a very active and multi-faceted field of study. What differentiates embedded computer vision from the general field of computer vision? Depending on the application, one will receive different answers. Is an embedded system simply a computer with monitor and keyboard removed? Or, is it a highly integrated system with specialized computing hardware? Does it perform simple image processing, or does it run highly complex recognition and even understanding algorithms? I was curious as to what definition this book would adopt.
At first it seemed my expectations were disappointed. On the surface the book is simply a collection of papers from different authors, each taking a different approach of the subject. Reading through yet another definition of and motivation for embedded vision systems started to get tiring. However, as we shall see this is not a bug but rather a feature.
The book is a result of the Embedded Computer Vision Workshop 2007. It was recognized that there is a lack of appropriate literature describing the current state of the art in this field. While the proceedings of the workshop do go a step in the right direction, it was felt that something more comprehensive is needed. Thus, the editors of the book asked the contributors to the workshop to write more extensive versions of the papers as chapters for the book. These chapters form the central Part II of the book. Part I sets the stage and serves as an introduction to the topic. It contains three chapters from different authors. Then, Part III looks ahead to what might come up in the future.
As each chapter was written by a different set of authors, and different aspects of embedded computer vision or different applications are covered. Each chapter starts again with its own introduction and motivation of the topic. This seems cumbersome at first. However, that the various authors do not always share the same view of embedded computer vision reflects the many facets of the field. This may sometimes force the reader to rethink their concept of embedded computer vision. For example, in Chapter 3 Alan J. Lipton writes about video analytics in a networked environment. At first it seems as though this topic should have no space in a book about embedded computer vision as video analytics is traditionally done on the server. However, the goal of the chapter is to show that more and more video analytics is done in the camera or in the network equipment to reduce network traffic. Thus, it becomes embedded.
The three chapters of Part I cover the broad spectrum of embedded computer vision. The first chapter focuses on hardware considerations and takes a look at the imaging sensors, the connection to the processor, and the processing hardware. Design methodology is the topic of the next chapter, and in the third and last chapter of Part I, video analytics in networks is covered.
Part II of the book contains seven chapters of which three present particularly efficient algorithms for use on resource-constrained platforms such as robust local features, motion history histograms , and multimodal mean background modeling. Other chapters are more hardware-centric and compare implementations of low level vision on DSPs, FPGAs, and mobile PC processors, show how to implement SAD-based stereo matching on FPGAs, make considerations for implementations on fixed-point DSPs, and present OpenVL to improve the real-time performance of computer vision applications.
The third part of the book is titled “Looking Ahead” and is comprised of three chapters which each present challenges for a particular set of applications. The first chapter of this part focuses on the particularly resource-constrained mobile applications. Another fast growing market is video analytics, the topic of the second chapter.
The book provides a very good overview of the current state of the art in embedded computer vision and of the major trends and growing markets. After reading the book one has not turned into an expert on embedded computer vision, but it is a good start and provides an extensive list of references to look for if one wants to go into more detail. Overall I would recommend this book to anyone interested in getting into this exciting field.
Embedded Computer Vision
Branislav Kisacanin, Shuvra S. Bhattacharyya,
and Sek Chai (Eds.)
Series: Advances in Computer Vision and Pattern Recognition
Marcus E. Hennecke (Austria)
Click on the image (above) to go to the publisher’s web page for this book where you will find a description of the book and the Table of Contents.