MOTION UNDERSTANDING Robot and Human Vision THE KLUWER INTERNATIONAL SERIES IN ENGINEERING AND COMPUTER SCIENCE ROBOTICS: VISION, MANIPULATION AND SENSORS Consulting Editor Takeo Kanade Carnegie Mellon University Other books in the series: Robotic Grasping and Fine Manipulation, M. Cutkosky ISBN 0-89838-200-9 Shadows and Silhouettes in Computer Vision, S. Shafer ISBN 0-89838-167-3 Perceptual Organization and Visual Recognition, D. Lowe ISBN 0-89838-172-X Robot Dynamics Algorithms, R. Featherstone ISBN 0-89838-230-0 Three Dimensional Machine Vision, T. Kanade, ed. ISBN 0-89838-188-6 Kinematic Modeling Identification, and Control of Robotic Manipulators, H.W. Stone ISBN 0-89838-237-8 Robotic Object Recognition Using Vision and Touch, P. Allen ISBN 0-89838-245-9 Integration, Coordination and Control of Multi-Sensor Robot Systems, H.F. Durrant-Whyte ISBN 0-89838-247-5 MOTION UNDERSTANDING Robot and Human Vision edited by W.N. Martin University of Virginia J.K. Aggarwal University of Texas at Austin ~. " KLUWER ACADEMIC PUBLISHERS Boston/Dordrecht/Lancaster Distributors for North America: Kluwer Academic Publishers 101 Philip Drive Assinippi Park Norwell, Massachusetts 02061 USA Distributors for the UK and Ireland: Kluwer Academic Publishers MTP Press Limited Falcon House, Queen Square Lancaster LAI lRN, UNITED KINGDOM Distributors for all other countries: Kluwer Academic Publishers Group Distribution Centre Post Office Box 322 3300 AH Dordrecht, THE NETHERLANDS Library of Congress Cataloging-in-Publication Data Motion understanding: robot and human vision / edited by W.N. Martin, J .K. Aggarwal. p. cm.-(The Kluwer international series in engineering and computer science; SECS44) Includes bibliographies and indexes. ISBN-13:978-1-4612-8413 -0 e-ISBN-13:978-1-4613 -1071-6 DOl: 10.1007/978-1-4613-1071-6 1. Robot vision. 2. Vision. 3. Motion. I. Martin, W.N. (Worthy N.) II. Aggarwal, J.K. III. Series. TJ211.3.M67 1988 87-31114 629.8'92-dcI9 CIP Copyright © 1988 by Kluwer Academic Publishers Softcover reprint of the hardcover 1st edition 1988 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, mechanical, photocopying, recording, or otherwise, without the prior written permission of the publisher, Kluwer Academic Publishers, 101 Philip Drive, Assinippi Park, Norwell, Massachusetts 02061. This book is dedicated to Shanti, Raj and Mala CONTENTS Contributors xiii Preface xvi Chapter 1 Bounding Constraint Propagation for Optical Flow Estimation Joseph K. Kearney and William B. Thompson 1.1 Introduction 1 1.2 The Gradient Constraint Equation 2 1.3 Gradient-Based Algorithms 3 1.4 Coping with Smoothness Violations 5 1.4.1 Thresholding for Smoothness 5 1.4.2 Continuous Adaptation to Errors 7 1.5 Results 12 1.6 Discussion 17 Chapter 2 Image Flow: Fundamentals and Algorithms Brian G. Schunck 2.1 Introduction 23 2.1.1 Background 24 2.1.2 Applications for Image Flow 26 2.1.3 Summary 30 2.2 Simple Image Flows 30 2.2.1 Image Flow Equation for Simple Flows 31 2.2.2 Algorithms for Simple Image Flows 32 2.2.3 Summary of Simple Image Flows 36 2.3 Discontinuous Image Flow 37 2.3.1 Surfaces and Projections 37 2.3.2 Image Irradiance Discontinuities 39 2.3.3 Velocity Field Discontinuities 41 2.3.4 Validity of the Image Flow Equation 42 2.3.5 Related Work 42 2.4 Analysis of Discontinuous Image Flows 44 2.4.1 Discontinuities in Continuous Image Functions 44 2.4.2 Sampling of Discontinuous Image Flows 48 2.4.3 Directional Selectivity 51 viii Contents 2.4.4 Summary of Discontinuous Image Flows 53 2.5 Algorithms for Discontinuous Image Flows 54 2.5.1 Background 54 2.5.2 Problem Statement 55 2.5.3 Constraint Line Clustering 56 2.5.4 Summary 60 2.6 Smoothing Discontinuous Image Flows 62 2.6.1 Motion Boundary Detection 63 2.6.2 Velocity Field Smoothing 64 2.6.3 Interleaved Detection and Smoothing 67 2.7 Summary and Conclusions 68 Chapter 3 A Computational Approach to the Fusion of Stereopsis and Kineopsis Amar Mitiche 3.1 Introduction 81 3.2 Integrating Optical Flow to Stereopsis for Motion 83 3.3 Perception of Rigid Objects in Motion 88 3.4 Examples 91 3.5 Summary 95 Chapter 4 The Empirical Study of Structure from Motion Myron L. Braunstein 4.1 Introduction 101 4.2 Viewer-Centered vs. Object-Centered Depth 102 4.2.1 Orthographic Projections of Rotation in Depth 106 4.2.2 Recovery of Structure from Velocity Gradients 110 4.3 The Correspondence Problem 113 4.3.1 Point Configurations 114 4.3.2 Contour Deformation 116 4.3.3 Texture Deformation 119 4.4 Rigidity 120 4.5 Perception of Self Motion 125 4.6 A Theory of Observers 127 4.7 An Empirical Test of Constraints 132 4.8 Summary and Conclusions 135 Contents ix Chapter 5 Motion Estimation Using More Than Two Images Hormoz Shariat and Keith Price 5.1 Introduction 143 5.2 General Description of the Method 146 5.2.1 Establishing the Equations 150 5.2.2 Simplifying the Equations 154 5.2.3 Solving the Equations 156 5.2.4 Calculating the Motion Parameters 157 5.2.5 Advantages of this Approach 159 5.2.6 Limitations of Our Approach 162 5.3 Results 163 5.3.1 Synthetic Test Data 164 5.3.2 Real Test Data 170 5.4 Comparison with Other Methods 182 5.4.1 Error Analysis 183 5.5 Conclusions 184 Chapter 6 An Experimental Investigation of Estimation Approaches for Optical Flow Fields W. Enkelmann, R. Kories, H.-H. Nagel and G. Zimmermann 6.1 Introduction 189 6.2 Feature Based Estimation 191 6.2.1 The Monotonicity Operator 191 6.2.2 From Feature Positions to Optical Flow Vectors 195 6.2.3 Test Sequence 195 6.2.4 Moving Object Detection 197 6.2.5 Performance Analysis of the Monotonicity Operator 199 6.2.6 Robustness of the Monotonicity Operator Against Parameter Changes 206 6.2.7 Reduction to Two Classes 208 6.3 Analytical Approach for the Estimation of Optical Flow Vector Fields 210 6.3.1 The "Oriented Smoothness" Constraint 211 6.3.2 Evaluation at Local Extrema of the Picture Function 216 6.4 Discussion 217 x Contents Chapter 7 The Incremental Rigidity Scheme and Long-Range Motion Correspondence Shimon Ullman 7.1 The Rigidity-Based Recovery of Structure from Motion 227 7.1.1 The Perception of Structure from Motion by Human Observers 227 7.1.2 Computational Studies of the Recovery of Structure from Motion 228 7.1.3 Additional Requirements for the Recovery of Structure from Motion 230 7.1.4 A Hypothesis: Maximizing Rigidity Relative to the Current Internal Model 232 7.2 The Incremental Rigidity Scheme 234 7.2.1 The Basic Scheme 235 7.2.2 Possible Modifications 238 7.2.3 Implementation 239 7.3 Experimental Results 242 7.3.1 Rigid Motion 242 7.3.2 Non-Rigid Motion 249 7.4 Additional Properties of the Incremental Rigidity Scheme 251 7.4.1 Orthographic and Perspective Projections 251 7.4.2 The Effect of the Number of Points 252 7.4.3 On Multiple Objects 256 7.4.4 Convergence to the Local Minimum 257 7.5 Possible Implications to the Long-Range Motion Correspondence Process 258 7.6 Summary 260 Chapter 8 Some Problems with Correspondence Michael Jenkin and Paul A. Kolers 8.1 Introduction 269 8.2 Determining Correspondence 275 8.3 Correspondence in Computer Vision 276 8.3.1 Correspondence in Stereopsis Algorithms 276 8.3.2 Correspondence in Temporal Matching Algorithms 278 8.4 An Experiment on Correspondence 282 8.5 Conclusions 289 Contents xi Chapter 9 Recovering Connectivity trom Moving Point-Light Displays Dennis R. Proffitt and Bennett I. Bertenthal 9.1 Introduction 297 9.2 Motion Information is a Minimal Stimulus Condition for the Perception of Form 299 9.3 Processing Models for Recovering Form from Motion 301 9.4 Do Fixed-Axis Models Predict Human Performance? 304 9.5 Human Implementation of Additional Processing Constraints 307 9.5.1 Centers of Moment 307 9.5.2 Occlusion's Effect on Depth Order and Implicit Form 309 9.5.3 Common Motion as Grouping Factor 314 9.5.4 Proximity 315 9.5.5 Familiarity 316 9.6 Incompatibilities Between Human Performance and Models Seeking Local Rigidity 319 9.6.1 Human Capabilities That Exceed Fixed-Axis Models: The Local Rigidity Assumption 319 9.6.2 Human Performance Limitations 320 9.7 Conclusion 321 Chapter 10 Algorithms for Motion Estimation Based on Three-Dimensional Correspondences S. D. Blostein and T. S. Huang 10.1 Introduction 329 10.2 Direct Linear Method 332 10.3 Method Based on Translation Invariants 333 10.4 Axis-Angle Method 336 10.5 The Screw Decomposition Method 338 10.6 Improved Motion Estimation Algorithms 343 10.7 Comparing the Linear and Nonlinear Methods 344 10.8 Simulation Results for Three-Point Methods 346 10.9 Some Recent Related Results 348
Description: