Vision-based Pedestrian Protection Systems for Intelligent Vehicles 基于视觉的智能驾驶系统.pdf

发布时间：2022-06-16 发布人：admin 分类：说明书资料大小：4.23M 资料格式：pdf 举报版权申诉

e8ffcf33-9d66-4469-a614-80af18e32297.pdf-第1页.png

第1页 / 共118页

e8ffcf33-9d66-4469-a614-80af18e32297.pdf-第2页.png

第2页 / 共118页

e8ffcf33-9d66-4469-a614-80af18e32297.pdf-第3页.png

第3页 / 共118页

e8ffcf33-9d66-4469-a614-80af18e32297.pdf-第4页.png

第4页 / 共118页

e8ffcf33-9d66-4469-a614-80af18e32297.pdf-第5页.png

第5页 / 共118页

e8ffcf33-9d66-4469-a614-80af18e32297.pdf-第6页.png

第6页 / 共118页

e8ffcf33-9d66-4469-a614-80af18e32297.pdf-第7页.png

第7页 / 共118页

e8ffcf33-9d66-4469-a614-80af18e32297.pdf-第8页.png

第8页 / 共118页

Preface

Contents

1 Introduction

1.1 Automobile's Impact

1.2 Advanced Driver Assistance Systems

1.3 Pedestrian Protection Systems

1.4 The Role of Computer Vision

1.5 Generic Framework

1.6 Book Outline

2 Candidates Generation

2.1 2D-Based Approaches

2.2 3D-Based Approaches

2.3 Motion-Based Approaches

2.4 Discussion

3 Classification

3.1 Preliminary Concepts

3.1.1 Image Descriptors

3.1.2 Pedestrian Models

3.1.3 Pedestrian Classifiers

3.2 Holistic Models: Focus on the Features

3.2.1 Templates

3.2.2 Haar Features

3.2.3 Edge Orientation Histograms Features

3.2.4 Histogram of Oriented Gradients Features

3.2.5 Shapelet Features

3.2.6 Local Binary Pattern Features

3.2.7 Dominant Orientation Template Features

3.2.8 Co-Occurrence Features

3.2.9 Covariance Features

3.2.10 Data-Driven Features

3.3 Diversified Models: From Features Fusion to Multiple Parts

3.3.1 Combined Features

3.3.2 Classifier Cascades/Ensembles

3.3.3 Multiple Aspects

3.3.4 Multiple (Body) Parts

3.3.5 Multiple Resolutions

3.3.6 Occlusion Handling

3.4 Training

3.4.1 Parameters Tuning

3.4.2 Bootstrapping

3.4.3 Data Annotation

3.4.4 Domain Adaptation

3.5 Discussion

4 Completing the System

4.1 Preprocessing

4.2 Verification and Refinement

4.3 Tracking

4.4 Application

4.5 Real-Time

4.6 Discussion

5 Datasets and Benchmarking

5.1 Datasets

5.2 Evaluation Protocols

5.3 Discussion

6 Conclusions

6.1 State of the Research

6.2 Future Challenges

References

SPRINGER BRIEFS IN COMPUTER SCIENCE David Gerónimo · Antonio M. López Vision-based Pedestrian Protection Systems for Intelligent Vehicles

SpringerBriefs in Computer Science Series Editors Stan Zdonik Peng Ning Shashi Shekhar Jonathan Katz Xindong Wu Lakhmi C. Jain David Padua Xuemin Shen Borko Furht V. S. Subrahmanian Martial Hebert Katsushi Ikeuchi Bruno Siciliano For further volumes: http://www.springer.com/series/10028

David Gerónimo • Antonio M. López Vision-based Pedestrian Protection Systems for Intelligent Vehicles 123

David Gerónimo Antonio M. López Computer Vision Center Universitat Autònoma de Barcelona Bellaterra, Barcelona Spain ISSN 2191-5768 ISBN 978-1-4614-7986-4 DOI 10.1007/978-1-4614-7987-1 Springer New York Heidelberg Dordrecht London ISSN 2191-5776 (electronic) ISBN 978-1-4614-7987-1 (eBook) Library of Congress Control Number: 2013940300 Ó The Author(s) 2014 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, speciﬁcally the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microﬁlms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection with reviews or scholarly analysis or material supplied speciﬁcally for the purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of this publication or parts thereof is permitted only under the provisions of the Copyright Law of the Publisher’s location, in its current version, and permission for use must always be obtained from Springer. Permissions for use may be obtained through RightsLink at the Copyright Clearance Center. Violations are liable to prosecution under the respective Copyright Law. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a speciﬁc statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. While the advice and information in this book are believed to be true and accurate at the date of publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for any errors or omissions that may be made. The publisher makes no warranty, express or implied, with respect to the material contained herein. Printed on acid-free paper Springer is part of Springer Science+Business Media (www.springer.com)

Preface i.e., This book is a natural evolution of the survey paper we published in the IEEE Transactions on Pattern Analysis and Machine Intelligence [121]. Therefore, we have followed the same point of view. On the one hand this means that we focus on vision-based solutions for pedestrian detection. Moreover, although we include some references from ﬁelds such as surveillance, robotics, and multimedia, the backbone context of the book is driver assistance, the development of pedestrian protection systems. On the other hand, we have organized the literature according to the main stages that a vision-based pedestrian protection system must incorporate. Moreover, we do not focus on the mathematical formalism behind each proposal. Rather, we try to explain the overall ideas so that interested readers can go to the original references to look into the details. Because of that, we think that novel researchers in this ﬁeld, from the academia or the industry, can take beneﬁt of this book. At this moment, [121] has received already more than 140 cites. We have informally collected feedback from some readers of [121] and we are glad to say that it seems that this paper accomplished its purpose, namely to give a quick overall organized idea of the ﬁeld to novel researchers and even to experienced researchers of other ﬁelds (e.g., it was interesting to discuss with researchers of the machine learning community about detection vs classiﬁcation). Thus, we hope that in the same line this book can be useful even for more researchers. In this book more than 300 papers are referenced, which gives an idea of the enormous interest that vision-based pedestrian detection has deserved from both the computer vision and the intelligent transportation systems communities. We include more than 190 references that were not included in [121], more than 80 published after [121], which shows how active this ﬁeld is. In spite of the amount of papers we reference, there are also many more that have not been included. We apologize for that, since all published works deserve their own acknowledgment given the effort of the authors. However, even a book like this one needs to limit the space in order to keep on focus. We do not provide a sort of best papers ranking, but we have chosen those papers that better allow us to illustrate the diversity of solutions available in the literature. Of course, this implies that papers that presented breaking ideas along this past decade are included. v

vi Preface It is worth to mention that vision-based pedestrian detection is a subﬁeld of object detection in videos, which in turn deserves its own book given the enormous corpus of proposals generated so far thanks to advances in topics such as machine learning, image descriptors, computational power, etc. This means that some general object detection methods can be applied to pedestrian detection. However, we have focused on the literature that explicitly addresses pedestrian detection in its own. As we will see, modern pedestrian/object detectors mainly rely on image descriptors (features), pedestrian/object models (e.g., holistic, multi-aspect, part- based, etc.), and learning machines (e.g., SVM and AdaBoost variants). In this book we focus our explanations on the content more speciﬁcally related to vision, i.e., pedestrian image descriptors and models. While the machine learning algo- rithms used in the reviewed literature are of course mentioned, explaining how such algorithms work is out of the scope of the book. This book is not done only thanks to the work of the authors, but many more persons and institutions must be acknowledged. First, we thank to the pedestrian detection community itself for addressing such a challenging and socially relevant problem. We thank also Springer for giving us the opportunity of writing this book. We thank to our daily collaborators at the Computer Vision Center (CVC) of Barcelona as well as at the Department de Ciències de la Computació of the Universitat Autònoma de Barcelona (UAB). Special thanks to the administrative and support staff who makes research easier. Many thanks to all the members of the CVC-ADAS group (www.cvc.uab.es/adas) for the hard work along the 10 years of life of the group and for being people that we really enjoy to work with. Special thanks also to David Váquez, Javier Marín, and Jiaolong Xu for kindly helping us to explain better some parts of this book. We want to give thanks also for the public funding we have received along the last years to support our research in the driver assistance context. In particular, we thank the following projects from the Spanish Government: TRA2011-29454-C03-01, TIN2011-29494-C03-02, TIN2011-25606, TRA2010-21371-C03-01, Consolider Ingenio 2010: MIPRCV (CSD2007-00018), TRA2007-62526/AUT, and TRA2004-06702/AUT. Antonio M. López also wants to thank his Ph.D. supervisor Joan Serrat, as well as mentors Juan José Villanueva, and Bart M. ter Haar Romeny. Good supervisors and mentors are key for having a successful career, these boys are. Equally important is to have good Ph.D. students who help you to stay updated and active in the frontier of the research. Accordingly, thanks also to Antonio’s Ph.D. stu- dents Daniel Ponsa, David Gerónimo, José M. Álvarez, David Vázquez, Javier Marín, Diego Cheda, Yainuvis Socarrás, Muhammad Rao, Jiaolong Xu, and Se- bastián Ramos. In addition, many thanks also to researchers who have hosted these students during researcher stages and with whom I have enjoyed research and life discussions. Thus, many thanks to Theo Gevers, Krystian Mikolajczyk, Ludmila Kuncheva, Dariu Gavrila, Bastian Leibe, and Frédéric Lerasle. Thanks also to other CVC-ADAS Ph.D. students with whom I have actively collaborated as Carme Julià, Ferrán Diego, José C. Rubio, and Germán Ros. As head of the CVC- ADAS team, Antonio also wants to give special thanks to the members who fought to build the team from the scratch, namely Joan Serrat, Felipe Lumbreras, Angel

Preface vii D. Sappa, and Daniel Ponsa. Antonio wants to thank also the private companies that trusted the CVC-ADAS team to develop some ADAS-related projects, in particular, thanks to Volkswagen A.G. at Wolfsburg (Thorsten Graf, Jörg Hilgenstock) and SEAT Centro Técnico of Martorell. Finally, Antonio wants to apologize with the family, specially with his parents, Antonio and Iluminada, his brother Juan, and specially with his wife Ana for stolen so many time supposed to be free-time for attending the research, many thanks for the enormous patience and support. David Gerónimo wants to thank his former Ph.D. supervisor and currently research collaborator Antonio M. López for the guidance and support during the last 8 years. Day by day, his attitude toward research has been an inspiring source for him to improve as a researcher. Many thanks also to his collaborators, research fellows and friends for the long and fruitful discussions: Angel D. Sappa, Joan Serrat, Daniel Ponsa, José M. Álvarez, David Vázquez, Javier Marín, Didier Devaurs, Hirofumi Uemura, Alejandro Mosteo, David Aldavert and Xuan Zou. Thanks also to David’s hosts during research stays in Surrey and LAAS-CNRS, Krystian Mikolajczyk and Frédéric Lerasle, for their warm welcome to their labs. Finally, David wants to thank his family, specially to his parents Angel and Núria, and his sister Sara, for their support. Finally, David’s most sincere acknowledg- ments are addressed to his wife Verónica, for her understanding and never-ending support, specially during the long and untimely hours spent in research.

Contents 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Automobile’s Impact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Advanced Driver Assistance Systems. . . . . . . . . . . . . . . . . . . . 1.3 Pedestrian Protection Systems . . . . . . . . . . . . . . . . . . . . . . . . . 1.4 The Role of Computer Vision . . . . . . . . . . . . . . . . . . . . . . . . . 1.5 Generic Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.6 Book Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Candidates Generation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2D-Based Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 2.2 3D-Based Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Motion-Based Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.1 3.1.2 3.1.3 3 Classiﬁcation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Preliminary Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Image Descriptors . . . . . . . . . . . . . . . . . . . . . . . . . . . Pedestrian Models . . . . . . . . . . . . . . . . . . . . . . . . . . . Pedestrian Classifiers . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Holistic Models: Focus on the Features . . . . . . . . . . . . . . . . . . Templates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 Haar Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.2 Edge Orientation Histograms Features . . . . . . . . . . . . . 3.2.3 Histogram of Oriented Gradients Features . . . . . . . . . . 3.2.4 Shapelet Features. . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.5 Local Binary Pattern Features . . . . . . . . . . . . . . . . . . . 3.2.6 Dominant Orientation Template Features . . . . . . . . . . . 3.2.7 Co-Occurrence Features . . . . . . . . . . . . . . . . . . . . . . . 3.2.8 3.2.9 Covariance Features. . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.10 Data-Driven Features . . . . . . . . . . . . . . . . . . . . . . . . . 1 1 4 5 7 8 9 13 13 17 20 20 23 24 24 25 26 26 26 30 32 33 36 36 38 39 42 43 ix

分享到：

赞收藏

资料库

Vision-based Pedestrian Protection Systems for Intelligent Vehicles 基于视觉的智能驾驶系统.pdf

相关推荐

行业

热门标签

最新资料