Human Identification Based on Gait is authored by three famous professors from three prestigious universities that are located on three different continents. The authors have extensive background not only in a wide variety of image processing and computer vision problems (such as remote sensing and texture analysis) but also in many different biometric technologies (such as ear, face, writer, and iris recognition). They have also worked on many vision problems related to motion processing and analysis, technologies that provide foundation for automatic gait recognition. Gait recognition is defined as recognition of people by the way they walk or run. The authors are pioneers in the field, a field that is one of the youngest and most computationally intensive among biometrics. They have written this monograph with real love and passion for the field.
The advantage of gait recognition is that the signal can be acquired covertly with video cameras that are already in use in many surveillance applications. Most other biometrics cannot be acquired covertly at a distance and usually require installation of new sensors. The disadvantage of gait recognition is that it is highly computationally intensive and requires expensive computational resources. Further, the technology has not yet matured to a point where the error rates would be acceptable in most realistic deployment scenarios. In the short introduction chapter, the authors briefly mention that the most likely usage will be in future government surveillance applications. Without dwelling very much on the applicability of the technology, the authors proceed quickly to technical matters that begin with Chapter 2 and extend all the way up to Chapter 8, the last chapter of this 187 page volume.
Before getting into the technical portion of the book, it is worthy to note that the authors have presented the bibliography, which is quite exhaustive, neither in alphabetical order nor in chronological order, but rather classified in topics which coincide with the sequence of the chapter topics. This style is quite useful in looking up a published paper on a specific topic. Considering the fact that an index is also provided, the authors have done a great job in making this book a standard reference on automatic gait recognition.
Chapter 2 provides historical background on gait research. Early research in gait analysis was performed in the field of medicine. The aim of the medical research was to classify components of gait for the treatment of pathologically abnormal patients. Results of some of this early work, for example, a conclusion that rotation angles for hip and knee are distinctive (i.e., provide inter-class variability) has later been used in some of the model-based automatic gait recognition algorithms. Some of the intra-class variability in gait has also been studied in these fields, for example, gait can be affected by footwear, carrying luggage, consuming alcohol, age, tight or loose clothing, and even mood and music. Studies performed in many other fields such as biomechanics, psychology, automated motion analysis, and podiatry, have provided basis and background for conducting research in automatic gait recognition. Many such studies are cited in the chapter.
It is pleasing to find that the authors discuss gait databases so early in the book, in Chapter 3. Due to variations in demographics, environmental conditions, cameras, backgrounds, deployment scenarios, etc., estimation of absolute performance of biometric systems has remained elusive. As a result, biometrics evaluations are necessarily comparative. A technology evaluation typically involves a comparison of various algorithms on a common database. The authors describe the databases and data collection protocols for two early databases and a few recent databases. The early databases from UCSD and Soton (University of Southampton) were small and easy (constant walking speed, special clothes, background, lighting, etc.). The recent databases are bigger, richer, and more difficult. These databases include UMD's surveillance data; NIST/USF's outdoor data imaging subjects at a distance; GaTech's data combining marker based motion analysis with video imagery; CMU's multi-view indoor data; CASIA's outdoor data and Southampton's data which combines ground truth indoor data (processed by broadcast techniques) with video of the same subjects walking in an outdoor scenario (for computer vision analysis). The databases are described nicely. However, if a reader were to obtain any of these databases, she needs to figure out how to get the database from its citation. It would have been valuable if the authors had provided all these databases at a single place, such as a DVD included with the book or links from the book's website.
Chapter 4 reviews some of the early approaches for automatic gait recognition. The objective of the earliest approaches was to produce a "proof-of-concept" (i.e., to show that gait has smaller intra-class variations than inter-class variations and therefore can be used as a biometric). The early approaches were somewhat simplistic (for example, processing video data to obtain silhouettes in the first step and deriving walking signatures to perform the recognition in the second step). Some of the early silhouette-based approaches derived gait signature from spatiotemporal (translation and time) pattern, optical flow distribution, or object-model characterization (using Eigen value decomposition, canonical analysis, or linear discriminant analysis). Model-based techniques first located high level features (e.g., thigh) and then performed a fourier analysis to estimate frequency spectrum of the periodic leg motion.
Chapter 5 is the longest chapter of the book, in fact more than twice as long as the next longest chapter, Chapter 6. Chapter 5 goes into the technical details of the popular silhouette-based approaches. Chapter 6 goes into technical details of less popular model-based approaches. Chapter 7 goes into technical details of approaches beyond those described in Chapters 5 and 6. In Chapter 5, the authors tabulate seminal silhouette-based approaches, categorizing them into moving shape, shape+motion, structural, and modeled. The authors provide an outline of the basis of these approaches and how they can be used for gait recognition. Chapter 5 is divided into three main sections: extending shape description to moving shapes; procrustes and spatiotemporal silhouette analysis; and modeling, matching, shape and kinematics. In each of these sections, the authors provide not only the technical description of the methods but also the performance achieved. The model-based approaches, described in Chapter 6, primarily try to model the movement of the torso and/or legs. The authors categorize these approaches into two main sections: planer human modeling and kinematics-based people tracking and recognition in 3D space. The authors discuss view invariant gait recognition algorithms and gait biometric fusion in Chapter 7. In all three chapters, they provide performance estimates of the discussed algorithms.
Chapters 5, 6, and 7 end rather abruptly but then all the discussions, future directions, and conclusions are provided at one place in Chapter 8. In Chapter 8, the authors recognize the fact that gait recognition is still in its infancy and most algorithms have been implemented for simplified and controlled conditions, e.g., no occlusion during human walking, relatively plain background, lack of view generality, etc. The authors recognize that while good recognition accuracy has been reported in some of the works, most existing algorithms have been tested on relatively small and non-realistic databases. Future research trends, directions, and open problems are discussed in this chapter. The authors conclude that a lot of opportunity for further research and development exists in the area of automatic gait recognition. This should excite and motivate interested students and researchers.
Human Identification Based on Gait
Mark S. Nixon, Tieniu Tan and Rama Chellappa
Springer, November 2005
Reviewed by: Salil Prabhakar
Human Identification Based on Gait is the fourth volume in Springer’s International Series on Biometrics.
Click above to go to the publisher’s web page where there is a description of the book and where you can view the Table of Contents and link to other books by these authors.
Book Reviews Published in
the IAPR Newsletter
Mathematics of Digital Images
by Stuart Hogan
Advances in Image and Video Segmentation
Graph-Theoretic Techniques for Web Content Mining
by Schenker, Bunke, Last and Kandel
Handbook of Mathematical Models in Computer Vision
by Paragios, Chen, and Faugeras (Editors)
The Geometry of Information Retrieval
by van Rijsbergen
Biometric Inverse Problems
by Yanushkevich, Stoica, Shmerko and Popel
Correlation Pattern Recognition
by Kumar, Mahalanobis, and Juday
Pattern Recognition 3rd Edition
by Theodoridis and Koutroumbas
Dictionary of Computer Vision and
by R.B. Fisher, et. Al
Kernel Methods for Pattern Analysis
by Shawe-Taylor and Cristianini
Machine Vision Books
CVonline: an overview
The Guide to Biometrics by Bolle, et al
Pattern Recognition Books
Jul. ‘04 [pdf]
Human Identification at a Distance:
University of Maryland (UMD)
Georgia Tech (GaTech)
Carnegie Mellon University (CMU)
Chinese Academy of Sciences (CASIA)
University of Southampton www.gait.ecs.soton.ac.uk/