Computational Intelligence in Automotive Applications Episode 1 Part 1 pptx

As if reflecting the general lack of consensus on what constitutes the field of CI, this volume illustrates automotive applications of not only neural and fuzzy computations1which are cons

Danil Prokhorov (Ed.)

Computational Intelligence in Automotive Applications

Studies in Computational Intelligence, Volume 132

Editor-in-chief

Prof Janusz Kacprzyk

Systems Research Institute

Polish Academy of Sciences

ul Newelska 6

01-447 Warsaw

Poland

E-mail: kacprzyk@ibspan.waw.pl

Further volumes of this series can be found on our

homepage: springer.com

Vol 111 David Elmakias (Ed.)

New Computational Methods in Power System Reliability,

2008

ISBN 978-3-540-77810-3

Vol 112 Edgar N Sanchez, Alma Y Alan´ıs and Alexander

G Loukianov

Discrete-Time High Order Neural Control: Trained with

Kalman Filtering, 2008

ISBN 978-3-540-78288-9

Vol 113 Gemma Bel-Enguix, M Dolores Jimenez-Lopez and

Carlos Mart´ın-Vide (Eds.)

New Developments in Formal Languages and Applications,

2008

ISBN 978-3-540-78290-2

Vol 114 Christian Blum, Maria Jos´e Blesa Aguilera, Andrea

Roli and Michael Sampels (Eds.)

Hybrid Metaheuristics, 2008

ISBN 978-3-540-78294-0

Vol 115 John Fulcher and Lakhmi C Jain (Eds.)

Computational Intelligence: A Compendium, 2008

ISBN 978-3-540-78292-6

Vol 116 Ying Liu, Aixin Sun, Han Tong Loh, Wen Feng Lu

and Ee-Peng Lim (Eds.)

Advances of Computational Intelligence in Industrial

Systems, 2008

ISBN 978-3-540-78296-4

Vol 117 Da Ruan, Frank Hardeman and Klaas van der Meer

(Eds.)

Intelligent Decision and Policy Making Support Systems,

2008

ISBN 978-3-540-78306-0

Vol 118 Tsau Young Lin, Ying Xie, Anita Wasilewska

and Churn-Jung Liau (Eds.)

Data Mining: Foundations and Practice, 2008

ISBN 978-3-540-78487-6

Vol 119 Slawomir Wiak, Andrzej Krawczyk and Ivo Dolezel

(Eds.)

Intelligent Computer Techniques in Applied

Electromagnetics, 2008

ISBN 978-3-540-78489-0

Vol 120 George A Tsihrintzis and Lakhmi C Jain (Eds.)

Multimedia Interactive Services in Intelligent Environments,

2008

Vol 121 Nadia Nedjah, Leandro dos Santos Coelho and Luiza de Macedo Mourelle (Eds.)

Quantum Inspired Intelligent Systems, 2008

ISBN 978-3-540-78531-6 Vol 122 Tomasz G Smolinski, Mariofanna G Milanova and Aboul-Ella Hassanien (Eds.)

Applications of Computational Intelligence in Biology, 2008

ISBN 978-3-540-78533-0 Vol 123 Shuichi Iwata, Yukio Ohsawa, Shusaku Tsumoto, Ning Zhong, Yong Shi and Lorenzo Magnani (Eds.)

Communications and Discoveries from Multidisciplinary Data, 2008

ISBN 978-3-540-78732-7 Vol 124 Ricardo Zavala Yoe (Ed.)

Modelling and Control of Dynamical Systems: Numerical Implementation in a Behavioral Framework, 2008

ISBN 978-3-540-78734-1 Vol 125 Larry Bull, Bernad´o-Mansilla Ester and John Holmes (Eds.)

Learning Classifier Systems in Data Mining, 2008

ISBN 978-3-540-78978-9 Vol 126 Oleg Okun and Giorgio Valentini (Eds.)

Supervised and Unsupervised Ensemble Methods and their Applications, 2008

ISBN 978-3-540-78980-2 Vol 127 R´egie Gras, Einoshin Suzuki, Fabrice Guillet and Filippo Spagnolo (Eds.)

Statistical Implicative Analysis, 2008

ISBN 978-3-540-78982-6 Vol 128 Fatos Xhafa and Ajith Abraham (Eds.)

Metaheuristics for Scheduling in Industrial and Manufacturing Applications, 2008

ISBN 978-3-540-78984-0 Vol 129 Natalio Krasnogor, Giuseppe Nicosia, Mario Pavone and David Pelta (Eds.)

Nature Inspired Cooperative Strategies for Optimization (NICSO 2007), 2008

ISBN 978-3-540-78986-4 Vol 130 Richi Nayak, N Ichalkaranje and Lakhmi C Jain (Eds.)

Evolution of Web in Artificial Intelligence Environments, 2008

ISBN 978-3-540-79139-3 Vol 131 Roger Lee and Haeng-Kon Kim (Eds.)

Computer and Information Science, 2008

ISBN 978-3-540-79186-7 Vol 132 Danil Prokhorov (Ed.)

Computational Intelligence in Automotive Applications, 2008

Danil Prokhorov

(Ed.)

Computational Intelligence

in Automotive Applications

With 157 Figures and 48 Tables

123

Danil Prokhorov

Toyota Technical Center - A Division

of Toyota Motor Engineering

and Manufacturing (TEMA)

Ann Arbor, MI 48105

USA

dvprokhorov@gmail.com

ISBN 978-3-540-79256-7 e-ISBN 978-3-540-79257-4

Studies in Computational Intelligence ISSN 1860-949X

Library of Congress Control Number: 2008925554

c

2008 Springer-Verlag Berlin Heidelberg

This work is subject to copyright All rights are reserved, whether the whole or part of the material is con-cerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilm or in any other way, and storage in data banks Duplication of this publica-tion or parts thereof is permitted only under the provisions of the German Copyright Law of September

9, 1965, in its current version, and permission for use must always be obtained from Springer-Verlag Violations are liable to prosecution under the German Copyright Law

The use of general descriptive names, registered names, trademarks, 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

Cover design: Deblik, Berlin, Germany

Printed on acid-free paper

9 8 7 6 5 4 3 2 1

springer.com

Computational Intelligence in Automotive Applications

What is computational intelligence (CI)? Traditionally, CI is understood as a collection of methods from the fields of neural networks (NN), fuzzy logic and evolutionary computation Various definitions and opinions exist, but what belongs to CI is still being debated; see, e.g., [1–3] More recently there has been a proposal

to define the CI not in terms of the tools but in terms of challenging problems to be solved [4]

With this edited volume I have made an attempt to give a representative sample of contemporary CI activities in automotive applications to illustrate the state of the art While CI research and achievements in some specialized fields described (see, e.g., [5, 6]), this is the first volume of its kind dedicated to automotive technology As if reflecting the general lack of consensus on what constitutes the field of CI, this volume illustrates automotive applications of not only neural and fuzzy computations1which are considered to be the “standard” CI topics, but also others, such as decision trees, graphical models, Support Vector Machines (SVM), multi-agent systems, etc

This book is neither an introductory text, nor a comprehensive overview of all CI research in this area Hopefully, as a broad and representative sample of CI activities in automotive applications, it will be worth reading for both professionals and students When the details appear insufficient, the reader is encouraged

to consult other relevant sources provided by the chapter authors

The chapter “Learning-based driver workload estimation” discusses research on estimation of driver cognitive workload and proposes a new methodology to design driver workload estimation systems The methodology is based on decision-tree learning It derives optimized models to assess the time-varying work-load level from data which include not only measurements from various sensors but also subjective workwork-load level ratings

The chapter “Visual monitoring of driver inattention” introduces a prototype computer vision system for real-time detection of driver fatigue The system includes an image acquisition module with an infrared illuminator, pupil detection and tracking module, and algorithms for detecting appropriate visual behaviors and monitoring six parameters which may characterize the fatigue level of a driver To increase effectiveness

of monitoring, a fuzzy classifier is implemented to fuse all these parameters into a single gauge of driver inattentiveness The system tested on real data from different drivers operates with high accuracy and robustly at night

The chapter “Understanding driving activity using ensemble methods” complements the chapter “Visual monitoring of driver inattention” by discussing whether driver inattention can be detected without eye and head tracking systems Instead of limiting themselves to working with just a few signals from preselected sensors, the authors chose to operate on hundreds of signals reflecting the real-time environment both outside and inside the vehicle The discovery of relationships in the data useful for driver activity classification, as

1Another “standard” CI topic called evolutionary computation (EC) is not represented in this volume in the form

of a separate chapter, although some EC elements are mentioned or referenced throughout the book Relevant publications on EC for automotive applications are available (e.g., [7]), but unfortunately were not available as

well as ranking signals in terms of their importance for classification, is entrusted to an approach called random forest, which turned out to be more effective than either hidden Markov models or SVM The chapter “Computer vision and machine learning for enhancing pedestrian safety” overviews methods for pedestrian detection, which use information from on-board and infrastructure based-sensors Many of the discussed methods are sufficiently generic to be useful for object detection, classification and motion prediction in general

The chapter “Application of graphical models in the automotive industry” describes briefly how graphical models, such as Bayesian and Markov networks, are used at Volkswagen and Daimler Production planning

at Volkswagen and demand prediction benefit significantly from the graphical model based system developed Another data mining system is developed for Daimler to help assessing the quality of vehicles and identifying causes of troubles when the vehicles have already spent some time in service It should be noted that other automotive companies are also pursuing data mining research (see, e.g., [8])

The chapter “Extraction of maximum support rules for the root cause analysis” discusses extraction of rules from manufacturing data for root cause analysis and process optimization An alternative approach to traditional methods of root cause analysis is proposed This new approach employs branch-and-bound princi-ples, and it associates process parameters with results of measurements, which is helpful in the identification

of the main drivers for quality variations of an automotive manufacturing process

The chapter “Neural networks in automotive applications” provides an overview of neural network tech-nology, concentrating on three main roles of neural networks: models, virtual or soft sensors and controllers Training of NN is also discussed, followed by a simple example illustrating the importance of recurrent NN The chapter “On learning machines for engine control” deals with modeling for control of turbocharged spark ignition engines with variable camshaft timing Two examples are considered: (1) estimation of the in-cylinder air mass in which open loop neural estimators are combined with a dynamic polytopic observer, and (2) modeling an in-cylinder residual gas fraction by a linear programming support vector regression method The authors argue that models based on first principles (“white boxes”) and neural or other “black box” models must be combined and utilized in the “grey box” approach to obtain results which are not just superior to any alternatives but are also more acceptable to automotive engineers

The chapter “Recurrent neural networks for AFR estimation and control in spark ignition automotive engines” complements the chapter “On learning machines for engine control” by discussing specifics of the air-fuel ratio (AFR) control Recurrent NN are trained off-line and employed as both the AFR virtual sensor and the inverse model controller The authors also provide a comparison with a conventional control strategy

on a real engine

The chapter “Intelligent vehicle power management: An overview” presents four case studies: a conven-tional vehicle power controller and three different approaches for a parallel HEV power controller They include controllers based on dynamic programming and neural networks, and fuzzy logic controllers, one of which incorporates predictions of driving environments and driving patterns

The chapter “Integrated diagnostic process for automotive systems” provides an overview of model-based and data-driven diagnostic methods applicable to complex systems Selected methods are applied to three automotive examples, one of them being a hardware-in-the-loop system, in which the methods are put to work together to solve diagnostic and prognostic problems It should be noted that integration of different approaches is an important theme for automotive research spanning the entire product life cycle (see, e.g., [9])

The chapter “Automotive manufacturing: intelligent resistance welding” introduces a real-time control system for resistance spot welding The control system is built on the basis of neural networks and fuzzy logic It includes a learning vector quantization NN for assessing the quality of weld nuggets and a fuzzy logic process controller Experimental results indicate substantial quality improvement over a conventional controller

The chapter “Intelligent control of mobility systems” (ICMS) overviews projects of the ICMS Program

at the National Institute of Standards and Technology (NIST) The program provides architecture, interface and data standards, performance test methods and infrastructure technology available to the manufacturing industry and government agencies in developing and applying intelligent control technology to mobility

systems (4D/RCS) control architecture for intelligent systems proposed and developed in the NIST Intelligent Systems Division

Unlike the book’s index, each chapter has its own bibliography for the convenience of the reader, with little overlap among references of different chapters

This volume highlights important challenges facing CI in the automotive domain Better vehicle diag-nostics/vehicle system safety, improved control of vehicular systems and manufacturing processes to save resources and minimize impact on the environment, better driver state monitoring, improved safety of pedes-trians, making vehicles more intelligent on the road – these are important directions where the CI technology can and should make the impact All of these are consistent with the Toyota vision [10]:

Toyota’s vision is to balance “Zeronize” and “Maximize” “Zeronize” symbolizes the vision and philosophy

of our persistent efforts in minimizing negative aspects vehicles have such as environmental impact, traffic congestion and traffic accidents, while “Maximize” symbolizes the vision and philosophy of our persistent efforts in maximizing the positive aspects vehicles have such as fun, delight, excitement and comfort, that people seek in automobiles.

I am very thankful to all the contributors of this edited volume for their willingness to participate in this project, their patience and valuable time I am also grateful to Prof Janusz Kacprzyk, the Springer Series Editor, for his encouragement to organize and edit this volume, as well as Thomas Ditzinger, the Springer production editor for his support of this project

January 2008

References

1 http://en.wikipedia.org/wiki/Computational intelligence

2 J.C Bezdek, “What is computational intelligence?” In Zurada, Marks and Robinson (Eds.), Computational

Intelligence: Imitating Life, pp 1–12, IEEE Press, New York, 1994.

3 R.J Marks II, “Intelligence: Computational Versus Artificial,” IEEE Transactions on Neural Networks, 4(5),

737–739, September, 1993

4 W Duch, “What is computational intelligence and what could it become?” In W Duch and J Mandziuk

(Eds.), Challenges for Computational Intelligence, Vol 63 of Studies in Computational Intelligence (J.

Kacprzyk Series Editor), Springer, Berlin Heidelberg New York, 2007 The chapter is available on-line at http://cogprints.org/5358/

5 Intelligent Control Systems Using Computational Intelligence Techniques (IEE Control Series) Edited by Antonio

Ruano, IEE, 2005

6 R Begg, Daniel T.H Lai, M Palaniswami Computational Intelligence in Biomedical Engineering CRC Press,

Taylor & Francis Books, Boca Raton, Florida, 2007

7 Marco Laumanns and Nando Laumanns, “Evolutionary Multiobjective Design in Automotive Development,”

Applied Intelligence, 23, 55–70, 2005.

8 T.A Montgomery, “Text Mining on a Budget: Reduce, Reuse, Recycle,” Michigan Leadership Summit on

Business Intelligence and Advanced Analytics, Troy, MI, March 8, 2007 Presentation is available on-line at

http://www.cmurc.com/bi-PreviousEvents.htm

9 P Struss and C Price, “Model-Based Systems in the Automotive Industry,” AI Magazine, Vol 24, No 4,

pp 17–34, AAAI, Menlo Park, Winter 2003

10 Toyota ITS vision, http://www.toyota.co.jp/en/tech/its/vision/

Learning-Based Driver Workload Estimation

Yilu Zhang, Yuri Owechko, and Jing Zhang 1

1 Background 1

2 Existing Practice and Its Challenges 3

3 The Proposed Approach: Learning-Based DWE 4

3.1 Learning-Based DWE Design Process 4

3.2 Benefits of Learning-Based DWE 5

4 Experimental Data 6

5 Experimental Process 8

6 Experimental Results 10

6.1 Driver-Independent Training 11

6.2 Driver-Dependent Training 13

6.3 Feature Combination 14

7 Conclusions and Future Work 15

References 16

Visual Monitoring of Driver Inattention Luis M Bergasa, Jes´ us Nuevo, Miguel A Sotelo, Rafael Barea, and Elena Lopez 19

1 Introduction 19

2 Previous Work 20

3 System Architecture 21

3.1 Image Acquisition System 22

3.2 Pupil Detection and Tracking 24

3.3 Visual Behaviors 26

3.4 Driver Monitoring 28

4 Experimental Results 30

4.1 Test Sequences 30

4.2 Parameter Measurement for One of the Test Sequences 30

4.3 Parameter Performance 31

5 Discussion 33

6 Conclusions and Future Work 35

References 36

Understanding Driving Activity Using Ensemble Methods

Kari Torkkola, Mike Gardner, Chris Schreiner, Keshu Zhang, Bob Leivian, Harry Zhang,

and John Summers 39

1 Introduction 39

2 Modeling Naturalistic Driving 40

3 Database Creation 41

3.1 Experiment Design 41

3.2 Annotation of the Database 42

4 Driving Data Classification 43

4.1 Decision Trees 44

4.2 Random Forests 45

4.3 Random Forests for Driving Maneuver Detection 46

5 Sensor Selection Using Random Forests 47

5.1 Sensor Selection Results 48

5.2 Sensor Selection Discussion 50

6 Driver Inattention Detection Through Intelligent Analysis of Readily Available Sensors 50

6.1 Driver Inattention 50

6.2 Inattention Data Processing 53

7 Conclusion 56

References 57

Computer Vision and Machine Learning for Enhancing Pedestrian Safety Tarak Gandhi and Mohan Manubhai Trivedi 59

1 Introduction 59

2 Framework for Pedestrian Protection System 60

3 Techniques in Pedestrian Detection 61

3.1 Candidate Generation 61

3.2 Candidate Validation 66

4 Infrastructure Based Systems 71

4.1 Background Subtraction and Shadow Suppression 71

4.2 Robust Multi-Camera Detection and Tracking 72

4.3 Analysis of Object Actions and Interactions 72

5 Pedestrian Path Prediction 72

6 Conclusion and Future Directions 75

References 76

Application of Graphical Models in the Automotive Industry Matthias Steinbrecher, Frank R¨ ugheimer, and Rudolf Kruse 79

1 Introduction 79

2 Graphical Models 80

2.1 Bayesian Networks 80

2.2 Markov Networks 80

3 Production Planning at Volkswagen Group 80

3.1 Data Description and Model Induction 81

3.2 Operations on the Model 82

3.3 Application 83

4 Vehicle Data Mining at Daimler AG 83

4.1 Data Description and Model Induction 84

4.2 Model Visualization 84

4.3 Application 85

5 Conclusion 88