Advanced Concepts for Intelligent

Vision Systems 19th International

Conference ACIVS 2018 Poitiers France
September 24 27 2018 Proceedings
Jacques Blanc-Talon
Visit to download the full and correct content document:
https://textbookfull.com/product/advanced-concepts-for-intelligent-vision-systems-19t
h-international-conference-acivs-2018-poitiers-france-september-24-27-2018-proceed
ings-jacques-blanc-talon/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...

Advanced Concepts for Intelligent Vision Systems 17th

International Conference ACIVS 2016 Lecce Italy October
24 27 2016 Proceedings 1st Edition Jacques Blanc-Talon

https://textbookfull.com/product/advanced-concepts-for-
intelligent-vision-systems-17th-international-conference-
acivs-2016-lecce-italy-october-24-27-2016-proceedings-1st-
edition-jacques-blanc-talon/

Advanced Concepts for Intelligent Vision Systems 18th

International Conference ACIVS 2017 Antwerp Belgium
September 18 21 2017 Proceedings 1st Edition Jacques
Blanc-Talon
https://textbookfull.com/product/advanced-concepts-for-
intelligent-vision-systems-18th-international-conference-
acivs-2017-antwerp-belgium-september-18-21-2017-proceedings-1st-
edition-jacques-blanc-talon/

Advanced Concepts for Intelligent Vision Systems 15th

International Conference ACIVS 2013 Pozna■ Poland
October 28 31 2013 Proceedings 1st Edition Christophe
Deknudt
https://textbookfull.com/product/advanced-concepts-for-
intelligent-vision-systems-15th-international-conference-
acivs-2013-poznan-poland-october-28-31-2013-proceedings-1st-
edition-christophe-deknudt/

Reachability Problems 12th International Conference RP

2018 Marseille France September 24 26 2018 Proceedings
Igor Potapov

https://textbookfull.com/product/reachability-problems-12th-
international-conference-rp-2018-marseille-france-
september-24-26-2018-proceedings-igor-potapov/
Knowledge and Systems Sciences 19th International
Symposium KSS 2018 Tokyo Japan November 25 27 2018
Proceedings Jian Chen

https://textbookfull.com/product/knowledge-and-systems-
sciences-19th-international-symposium-kss-2018-tokyo-japan-
november-25-27-2018-proceedings-jian-chen/

Computer Information Systems and Industrial Management:

17th International Conference, CISIM 2018, Olomouc,
Czech Republic, September 27-29, 2018, Proceedings
Khalid Saeed
https://textbookfull.com/product/computer-information-systems-
and-industrial-management-17th-international-conference-
cisim-2018-olomouc-czech-republic-
september-27-29-2018-proceedings-khalid-saeed/

Computer Vision and Graphics International Conference

ICCVG 2018 Warsaw Poland September 17 19 2018
Proceedings Leszek J. Chmielewski

https://textbookfull.com/product/computer-vision-and-graphics-
international-conference-iccvg-2018-warsaw-poland-
september-17-19-2018-proceedings-leszek-j-chmielewski/

Database Systems for Advanced Applications 25th

International Conference DASFAA 2020 Jeju South Korea
September 24 27 2020 Proceedings Part I Yunmook Nah

https://textbookfull.com/product/database-systems-for-advanced-
applications-25th-international-conference-dasfaa-2020-jeju-
south-korea-september-24-27-2020-proceedings-part-i-yunmook-nah/

Database Systems for Advanced Applications 25th

International Conference DASFAA 2020 Jeju South Korea
September 24 27 2020 Proceedings Part II Yunmook Nah

https://textbookfull.com/product/database-systems-for-advanced-
applications-25th-international-conference-dasfaa-2020-jeju-
south-korea-september-24-27-2020-proceedings-part-ii-yunmook-nah/
 Jacques Blanc-Talon · David Helbert
Wilfried Philips · Dan Popescu
Paul Scheunders (Eds.)

Advanced Concepts
LNCS 11182

for Intelligent
Vision Systems
19th International Conference, ACIVS 2018
Poitiers, France, September 24–27, 2018
Proceedings

123
Lecture Notes in Computer Science 11182
Commenced Publication in 1973
Founding and Former Series Editors:
Gerhard Goos, Juris Hartmanis, and Jan van Leeuwen

Editorial Board
David Hutchison
Lancaster University, Lancaster, UK
Takeo Kanade
Carnegie Mellon University, Pittsburgh, PA, USA
Josef Kittler
University of Surrey, Guildford, UK
Jon M. Kleinberg
Cornell University, Ithaca, NY, USA
Friedemann Mattern
ETH Zurich, Zurich, Switzerland
John C. Mitchell
Stanford University, Stanford, CA, USA
Moni Naor
Weizmann Institute of Science, Rehovot, Israel
C. Pandu Rangan
Indian Institute of Technology Madras, Chennai, India
Bernhard Steffen
TU Dortmund University, Dortmund, Germany
Demetri Terzopoulos
University of California, Los Angeles, CA, USA
Doug Tygar
University of California, Berkeley, CA, USA
Gerhard Weikum
Max Planck Institute for Informatics, Saarbrücken, Germany
More information about this series at http://www.springer.com/series/7412
Jacques Blanc-Talon David Helbert
•

Wilfried Philips Dan Popescu

•

Paul Scheunders (Eds.)

Advanced Concepts
for Intelligent
Vision Systems
19th International Conference, ACIVS 2018
Poitiers, France, September 24–27, 2018
Proceedings

123
Editors
Jacques Blanc-Talon Dan Popescu
DGA CSIRO-ICT Centre
Bagneux Canberra, ACT
France Australia
David Helbert Paul Scheunders
Laboratoire XLIM University of Antwerp
Futuroscope Chasseneuil Cedex Wilrijk
France Belgium
Wilfried Philips
Ghent University
Ghent
Belgium

ISSN 0302-9743 ISSN 1611-3349 (electronic)

Lecture Notes in Computer Science
ISBN 978-3-030-01448-3 ISBN 978-3-030-01449-0 (eBook)
https://doi.org/10.1007/978-3-030-01449-0

Library of Congress Control Number: 2018955578

LNCS Sublibrary: SL6 – Image Processing, Computer Vision, Pattern Recognition, and Graphics

© Springer Nature Switzerland AG 2018

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.
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.
The publisher, the authors and the editors are safe to assume that the advice and information in this book are
believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors
give a warranty, express or implied, with respect to the material contained herein or for any errors or
omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
 Preface

These proceedings gather the selected papers of the Advanced Concepts for Intelligent
Vision Systems (ACIVS) Conference, which was held in Poitiers, France, during
September 24–27, 2018.
This event was the 19th ACIVS. Since the first event in Germany in 1999, ACIVS
has become a larger and independent scientific conference. However, the seminal
distinctive governance rules have been maintained:
– To update the conference scope on a yearly basis. While keeping a technical
backbone (the classic low-level image processing techniques), we have introduced
topics of interest such as video analysis, segmentation, classification, remote
sensing, biometrics, deep learning, and image and video compression, restoration
and reconstruction. In addition, speakers usually give invited talks on hot issues.
– To remain a single-track conference in order to promote scientific exchanges among
the audience.
– To grant oral presentations a duration of 25 minutes and published papers a length
of 12 pages, which is significantly different from most other conferences.
The second and third items entail a complex management of the conference; in
particular, the number of time slots is rather small. Although the selection between the
two presentation formats is primarily determined by the need to compose a
well-balanced program, papers presented during plenary and poster sessions enjoy the
same importance and publication format.
The first item is strengthened by the notoriety of ACIVS, which has been growing
over the years: official Springer records show a cumulated number of downloads on
August 1, 2018, of more than 550,000 (for ACIVS 2005–2016 only).
The regular sessions also included a couple of invited talks by Yuliya Tarabalka
(Inria Sophia Antipolis, France) and Mihai Datcu (DLR, Germany). We would like to
thank all of them for enhancing the technical program with their presentations.
ACIVS attracted submissions from many different countries, mostly from Europe,
but also from the rest of the world: Belgium, China, Czech Republic, Finland, France,
Germany, India, Italy, Portugal, Korea, Romania, Saudi Arabia, Spain, the UK, the
USA, and Vietnam.
From 91 submissions, 36 were selected for oral presentation and 16 as posters. The
paper submission and review procedure was carried out electronically and a minimum
of two reviewers were assigned to each paper. A large and energetic Program Com-
mittee, helped by additional reviewers, as listed on the following pages, completed the
long and demanding reviewing process. We would like to thank all of them for their
timely and high-quality reviews, achieved in quite a short time and during the summer
holidays.
Finally, we would like to thank all the participants who trusted in our ability to
organize this conference for the 19th time. We hope they attended a different and
VI Preface

stimulating scientific event and that they enjoyed the atmosphere of the various ACIVS
social events in the city of Poitiers.
As mentioned, a conference like ACIVS would not be feasible without the concerted
effort of many people and the support of various institutions. We are indebted to the
local organizers Pascal Bourdon, David Helbert, Mohamed-Chaker Larabi, François
Lecellier, Benoit Tremblais, and Thierry Urruty, for having smoothed all the harsh
practical details of an event venue.

July 2018 Jacques Blanc-Talon

David Helbert
Wilfried Philips
Dan Popescu
Paul Scheunders
 Organization

Acivs 2018 was organized by the University of Antwerp, Belgium.

Steering Committee
Jacques Blanc-Talon DGA, France
David Helbert University of Poitiers, France
Wilfried Philips Ghent University - imec, Belgium
Dan Popescu CSIRO Data 61, Australia
Paul Scheunders University of Antwerp, Belgium

Organizing Committee
Pascal Bourdon XLIM, France
David Helbert University of Poitiers, France
Mohamed-Chaker Larabi XLIM, France
François Lecellier XLIM, France
Benoit Tremblais XLIM, France
Thierry Urruty XLIM, France

Program Committee
Hojjat Adeli Ohio State University, USA
Syed Afaq Shah The University of Western Australia, Australia
Hamid Aghajan Ghent University - imec, Belgium
Edoardo Ardizzone University of Palermo, Italy
Antonis Argyros University of Crete, Greece
George Bebis University of Nevada, USA
Fabio Bellavia University of Florence, Italy
Jenny Benois-Pineau University of Bordeaux, France
Dominique Béréziat Université Pierre et Marie Curie, France
Yannick Berthoumieu Bordeaux INP, France
Janus Bobulski Czestochowa University of Technology, Poland
Philippe Bolon University of Savoie, France
Egor Bondarev Technische Universiteit Eindhoven, The Netherlands
Don Bone University of Technology Sydney, Australia
Adrian Bors University of York, UK
Salah Bourennane Ecole Centrale de Marseille, France
Catarina Brites Instituto Superior Técnico, Portugal
Vittoria Bruni University of Rome La Sapienza, Italy
Dumitru Burdescu University of Craiova, Romania
VIII Organization

Tiago J. Carvalho Instituto Federal de São Paulo - Campinas, Brazil

Giuseppe Cattaneo University of Salerno, Italy
Jocelyn Chanussot Université de Grenoble Alpes, France
Kacem Chehdi ENSSAT, France
Gianluigi Ciocca University of Milano Bicocca, Italy
Eric Debreuve CNRS, France
Patrick Delmas The University of Auckland, New Zealand
Stéphane Derrode Ecole Centrale de Lyon, France
Nicolas Dobigeon ENSEEIHT, France
Jérôme Gilles San Diego State University, USA
Daniele Giusto Università degli Studi di Cagliari, Italy
Bart Goossens Ghent University - imec, Belgium
Philippe Gosselin ENSEA, France
Artyom Grigoryan UTSA, USA
Christine Guillemot Inria, France
Jon Yngve Hardeberg Norwegian University of Science and Technology,
Norway
Gloria Haro Universitat Pompeu Fabra, Spain
Monson Hayes George Mason University, USA
David Helbert University of Poitiers, France
Lionel Heng DSO National Laboratories, Singapore
Michael Hild Osaka Electro-Communication University, Japan
Mark Holden Kyoto University, Japan
Kazuhiro Hotta Meijo University, Japan
Dimitris Iakovidis University of Thessaly, Greece
Syed Islam Edith Cowan University, Australia
Yuji Iwahori Chubu University, Japan
Graeme Jones Kingston University, UK
Arto Kaarna Lappeenranta University of Technology, Finland
Robert Koprowski University of Silesia, Poland
Bogdan Kwolek AGH University of Science and Technology, Poland
Kenneth Lam The Hong Kong Polytechnic University, SAR China
Patrick Lambert Polytech’ Savoie, France
Sébastien Lefèvre Université Bretagne Sud, France
Ludovic Macaire Université Lille 1 Sciences et Technologies, France
Antoine Manzanera ENSTA ParisTech, France
Gonzalo Pajares Martinsanz Universidad Complutense, Spain
Fabrice Mériaudeau Universiti Teknologi PETRONAS, Malaysia
Massimo Minervini IMT School for Advanced Studies, Italy
Amar Mitiche INRS, Canada
David Monnin French–German Research Institute of Saint-Louis,
France
Adrian Munteanu Vrije Universiteit Brussel, Belgium
Maria Navascuès Universidad de Zaragoza, Spain
António J. R. Neves University of Aveiro, Portugal
Jennifer Newman Iowa State University, USA
 Organization IX

Vincent Nozick Université Paris-Est Marne-la-Vallée, France

Danielle Nuzillard Université de Reims Champagne-Ardenne, France
Rudi Penne University of Antwerp, Belgium
Fernando Pérez-González University of Vigo, Spain
Caroline Petitjean Université de Rouen, France
Hossein Rahmani Lancaster University, UK
Giovanni Ramponi University of Trieste, Italy
Florent Retraint Université de Technologie de Troyes, France
Patrice Rondao Alface Nokia Bell Labs, Belgium
Florence Rossant ISEP, France
Luis Salgado Universidad Politécnica, Spain
Nel Samama Télécom Sud-Paris, France
Ivan Selesnick New York University, USA
Wladyslaw Skarbek University of Technology, Poland
Andrzej Sluzek Khalifa University, United Arab Emirates
Ferdous Sohel Murdoch University, Australia
Changming Sun CSIRO, Australia
Hugues Talbot Université Paris-Est - ESIEE, France
Attila Tanács University of Szeged, Hungary
Yuliya Tarabalka Inria, France
Nadège Thirion-Moreau SeaTech - Université de Toulon, France
Sylvie Treuillet Université d’Orléans, France
Florence Tupin Télécom ParisTech, Université Paris Saclay, France
Cesare Valenti Università di Palermo, Italy
Marc Van Droogenbroeck University of Liège, Belgium
Pascal Vasseur LITIS - Université de Rouen Normandie, France
Peter Veelaert Ghent University - imec, Belgium
Sergio Velastin Queen Mary University of London, UK
Nicole Vincent Université Paris Descartes, France
Domenico Vitulano National Research Council, Italy
Damien Vivet ISAE-SUPAERO, France
Shin Yoshizawa RIKEN, Japan
Gerald Zauner Fachhochschule Ober Osterreich, Austria
Pavel Zemcik Brno University of Technology, Czech Republic
Josiane Zérubia Inria, France
Djemel Ziou Sherbrooke University, Canada

Additional Reviewers
Syed Afaq Shah The University of Western Australia, Australia
Hamid Aghajan Ghent University - imec, Belgium
Roxana Agrigoroaie U2IS, ENSTA ParisTech, France
Edoardo Ardizzone University of Palermo, Italy
Antonis Argyros University of Crete, Greece
George Bebis University of Nevada, USA
Fabio Bellavia University of Florence, Italy
X Organization

Yannick Berthoumieu Bordeaux INP, France

Jacques Blanc-Talon DGA, France
Janus Bobulski Czestochowa University of Technology, Poland
Philippe Bolon University of Savoie, France
Egor Bondarev Technische Universiteit Eindhoven, The Netherlands
Don Bone University of Technology Sydney, Australia
Adrian Bors University of York, UK
Salah Bourennane Ecole Centrale de Marseille, France
Catarina Brites Instituto Superior Técnico, Portugal
Vittoria Bruni University of Rome La Sapienza, Italy
Dumitru Burdescu University of Craiova, Romania
Tiago J. Carvalho Instituto Federal de São Paulo - Campinas, Brazil
Giuseppe Cattaneo University of Salerno, Italy
Jocelyn Chanussot Université de Grenoble Alpes, France
Kacem Chehdi ENSSAT, France
Gianluigi Ciocca University of Milano Bicocca, Italy
Eric Debreuve CNRS, France
Jérôme Gilles San Diego State University, USA
Daniele Giusto Università degli Studi di Cagliari, Italy
Christine Guillemot Inria, France
Jon Yngve Hardeberg Norwegian University of Science and Technology,
Norway
Gloria Haro Universitat Pompeu Fabra, Spain
Monson Hayes George Mason University, USA
David Helbert University of Poitiers, France
Lionel Heng DSO National Laboratories, Singapore
Michael Hild Osaka Electro-Communication University, Japan
Mark Holden Kyoto University, Japan
Kazuhiro Hotta Meijo University, Japan
Dimitris Iakovidis University of Thessaly, Greece
Yuji Iwahori Chubu University, Japan
Arto Kaarna Lappeenranta University of Technology, Finland
Bogdan Kwolek AGH University of Science and Technology, Poland
Sébastien Lefèvre Université Bretagne Sud, France
Ludovic Macaire Université Lille 1 Sciences et Technologies, France
Antoine Manzanera ENSTA ParisTech, France
Gonzalo Pajares Martinsanz Universidad Complutense de Madrid, Spain
Adrian Munteanu Vrije Universiteit Brussel, Belgium
António J. R. Neves University of Aveiro, Portugal
Rudi Penne University of Antwerp, Belgium
Wilfried Philips Ghent University - imec, Belgium
Clement Pinard ENSTA Paristech, France
Ljiljana Platisa Ghent University - imec, Belgium
Dan Popescu CSIRO Data 61, Australia
Hossein Rahmani Lancaster University, UK
Giovanni Ramponi University of Trieste, Italy
 Organization XI

Patrice Rondao Alface Nokia Bell Labs, Belgium

Florence Rossant ISEP, France
Luis Salgado Universidad Politécnica, Spain
Paul Scheunders University of Antwerp, Belgium
Ivan Selesnick New York University, USA
Wladyslaw Skarbek University of Technology, Poland
Andrzej Sluzek Khalifa University, United Arab Emirates
Ferdous Sohel Murdoch University, Australia
Changming Sun CSIRO, Australia
Attila Tanács University of Szeged, Hungary
Yuliya Tarabalka Inria, France
Sylvie Treuillet Université d’Orléans, France
Pascal Vasseur LITIS - Université de Rouen Normandie, France
Peter Veelaert Ghent University - imec, Belgium
Sergio Velastin Queen Mary University of London, UK
Nicole Vincent Université Paris Descartes, France
Damien Vivet ISAE-SUPAERO, France
Shin Yoshizawa RIKEN, Japan
Gerald Zauner Fachhochschule Ober Osterreich, Austria
Pavel Zemcik Brno University of Technology, Czech Republic
Josiane Zérubia Inria, France
 Contents

Video Analysis

Improving a Switched Vector Field Model for Pedestrian

Motion Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Catarina Barata, Jacinto C. Nascimento, and Jorge S. Marques

Matrix Descriptor of Changes (MDC): Activity Recognition

Based on Skeleton. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Radek Simkanič

Person Re-Identification with a Body Orientation-Specific Convolutional

Neural Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Yiqiang Chen, Stefan Duffner, Andrei Stoian, Jean-Yves Dufour,
and Atilla Baskurt

Distributed Estimation of Vector Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

Ana Portêlo, Jorge S. Marques, Catarina Barata, and João M. Lemos

Clustering Based Reference Normal Pose for Improved

Expression Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Andrei Racoviţeanu, Iulian Felea, Laura Florea, Mihai Badea,
and Corneliu Florea

Detecting and Recognizing Salient Object in Videos . . . . . . . . . . . . . . . . . . 62

Rahma Kalboussi, Mehrez Abdellaoui, and Ali Douik

Directional Beams of Dense Trajectories for Dynamic

Texture Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
Thanh Tuan Nguyen, Thanh Phuong Nguyen, Frédéric Bouchara,
and Xuan Son Nguyen

Intrinsic Calibration of a Camera to a Line-Structured Light Using a Single

View of Two Spheres . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Yu Liu, Jiayu Yang, Xiaoyong Zhou, Qingqing Ma, and Hui Zhang

3D Object-Camera and 3D Face-Camera Pose Estimation for Quadcopter

Control: Application to Remote Labs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Fawzi Khattar, Fadi Dornaika, Benoit Larroque, and Franck Luthon

Orthogonally-Divergent Fisheye Stereo . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

Janice Pan, Martin Mueller, Tarek Lahlou, and Alan C. Bovik
XIV Contents

Two-Camera Synchronization and Trajectory Reconstruction for a Touch

Screen Usability Experiment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Toni Kuronen, Tuomas Eerola, Lasse Lensu, and Heikki Kälviäinen

Segmentation and Classification

Comparison of Co-segmentation Methods for Wildlife Photo-identification. . . 139

Anastasia Popova, Tuomas Eerola, and Heikki Kälviäinen

An Efficient Agglomerative Algorithm Cooperating with Louvain Method

for Implementing Image Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
Thanh-Khoa Nguyen, Mickael Coustaty, and Jean-Loup Guillaume

Robust Feature Descriptors for Object Segmentation Using Active

Shape Models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
Daniela Medley, Carlos Santiago, and Jacinto C. Nascimento

Foreground Background Segmentation in Front of Changing Footage

on a Video Screen. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
Gianni Allebosch, Maarten Slembrouck, Sanne Roegiers,
Hiêp Quang Luong, Peter Veelaert, and Wilfried Philips

Multi-organ Segmentation of Chest CT Images in Radiation Oncology:

Comparison of Standard and Dilated UNet . . . . . . . . . . . . . . . . . . . . . . . . . 188
Umair Javaid, Damien Dasnoy, and John A. Lee

Diffuse Low Grade Glioma NMR Assessment for Better Intra-operative

Targeting Using Fuzzy Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
Mathieu Naudin, Benoit Tremblais, Carole Guillevin, Rémy Guillevin,
and Christine Fernandez-Maloigne

Identification of Saimaa Ringed Seal Individuals Using Transfer Learning . . . 211

Ekaterina Nepovinnykh, Tuomas Eerola, Heikki Kälviäinen,
and Gleb Radchenko

Remote Sensing

Enhanced Codebook Model and Fusion for Object Detection with

Multispectral Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
Rongrong Liu, Yassine Ruichek, and Mohammed El Bagdouri

Unsupervised Perception Model for UAVs Landing Target Detection

and Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
Eric Bazán, Petr Dokládal, and Eva Dokládalová
 Contents XV

Parallel and Distributed Local Fisher Discriminant Analysis to Reduce

Hyperspectral Images on Cloud Computing Architectures. . . . . . . . . . . . . . . 245
Rania Zaatour, Sonia Bouzidi, and Ezzeddine Zagrouba

Bayesian Vehicle Detection Using Optical Remote Sensing Images . . . . . . . . 258

Walma Gharbi, Lotfi Chaari, and Amel Benazza-Benyahia

Integrating UAV in IoT for RoI Classification in Remote Images . . . . . . . . . 270

Loretta Ichim and Dan Popescu

Biometrics

Enhanced Line Local Binary Patterns (EL-LBP): An Efficient Image

Representation for Face Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
Hung Phuoc Truong and Yong-Guk Kim

Single Sample Face Recognition by Sparse Recovery of Deep-Learned

LDA Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
Matteo Bodini, Alessandro D’Amelio, Giuliano Grossi,
Raffaella Lanzarotti, and Jianyi Lin

Recursive Chaining of Reversible Image-to-Image Translators

for Face Aging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
Ari Heljakka, Arno Solin, and Juho Kannala

Automatically Selecting the Best Pictures for an Individualized Child

Photo Album . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
Floris De Feyter, Kristof Van Beeck, and Toon Goedemé

Face Detection in Painting Using Deep Convolutional Neural Networks . . . . 333

Olfa Mzoughi, André Bigand, and Christophe Renaud

Robust Geodesic Skeleton Estimation from Body Single Depth. . . . . . . . . . . 342

Jaehwan Kim and Howon Kim

Deep Learning

Analysis of Neural Codes for Near-Duplicate Detection . . . . . . . . . . . . . . . . 357

Maurizio Pintus

Optimum Network/Framework Selection from High-Level Specifications

in Embedded Deep Learning Vision Applications . . . . . . . . . . . . . . . . . . . . 369
Delia Velasco-Montero, Jorge Fernández-Berni,
Ricardo Carmona-Galán, and Ángel Rodríguez-Vázquez
XVI Contents

Contour Propagation in CT Scans with Convolutional Neural Networks. . . . . 380

Jean Léger, Eliott Brion, Umair Javaid, John Lee,
Christophe De Vleeschouwer, and Benoit Macq

Person Re-identification Using Group Context . . . . . . . . . . . . . . . . . . . . . . 392

Yiqiang Chen, Stefan Duffner, Andrei Stoian, Jean-Yves Dufour,
and Atilla Baskurt

Fingerprint Classification Using Conic Radon Transform and

Convolutional Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402
Dhekra El Hamdi, Ines Elouedi, Abir Fathallah, Mai K. Nguyen,
and Atef Hamouda

NoiseNet: Signal-Dependent Noise Variance Estimation with

Convolutional Neural Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414
Mykhail Uss, Benoit Vozel, Vladimir Lukin, and Kacem Chehdi

Effective Training of Convolutional Neural Networks for Insect

Image Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
Maxime Martineau, Romain Raveaux, Clément Chatelain,
Donatello Conte, and Gilles Venturini

A Deep Learning Approach to Hair Segmentation and Color Extraction

from Facial Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 438
Diana Borza, Tudor Ileni, and Adrian Darabant

Learning Morphological Operators for Depth Completion. . . . . . . . . . . . . . . 450

Martin Dimitrievski, Peter Veelaert, and Wilfried Philips

Dealing with Topological Information Within a Fully Convolutional

Neural Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 462
Etienne Decencière, Santiago Velasco-Forero, Fu Min, Juanjuan Chen,
Hélène Burdin, Gervais Gauthier, Bruno Laÿ, Thomas Bornschloegl,
and Thérèse Baldeweck

Coding and Compression

L-Infinite Predictive Coding of Depth . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475

Wenqi Chang, Ionut Schiopu, and Adrian Munteanu

An Application of Data Compression Models to Handwritten

Digit Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 487
Armando J. Pinho and Diogo Pratas

A Global Decoding Strategy with a Reduced-Reference Metric Designed

for the Wireless Transmission of JPWL . . . . . . . . . . . . . . . . . . . . . . . . . . . 496
Xinwen Xie, Philippe Carré, Clency Perrine, Yannis Pousset,
Jianhua Wu, and Nanrun Zhou
 Contents XVII

Reconfigurable FPGA Implementation of the AVC Quantiser

and De-quantiser Blocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 506
Vijaykumar Guddad, Amit Kulkarni, and Dirk Stroobandt

Image Restoration and Reconstruction

Large Parallax Image Stitching Using an Edge-Preserving Diffeomorphic

Warping Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 521
Geethu Miriam Jacob and Sukhendu Das

A Wavelet Based Image Fusion Method Using Local Multiscale

Image Regularity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534
Vittoria Bruni, Alessandra Salvi, and Domenico Vitulano

Optimising Data for Exemplar-Based Inpainting . . . . . . . . . . . . . . . . . . . . . 547

Lena Karos, Pinak Bheed, Pascal Peter, and Joachim Weickert

Fast Light Field Inpainting Propagation Using Angular Warping

and Color-Guided Disparity Interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . 559
Pierre Allain, Laurent Guillo, and Christine Guillemot

Fusing Omnidirectional Visual Data for Probability Matching Prediction . . . . 571

David Valiente, Luis Payá, Luis M. Jiménez, Jose M. Sebastián,
and Oscar Reinoso

Derivative Half Gaussian Kernels and Shock Filter . . . . . . . . . . . . . . . . . . . 584

Baptiste Magnier, Vincent Noblet, Adrien Voisin, and Dylan Legouestre

Scanner Model Identification of Official Documents Using Noise

Parameters Estimation in the Wavelet Domain . . . . . . . . . . . . . . . . . . . . . . 598
Chaima Ben Rabah, Gouenou Coatrieux, and Riadh Abdelfattah

Relocated Colour Contrast Occurrence Matrix and Adapted Similarity

Measure for Colour Texture Retrieval . . . . . . . . . . . . . . . . . . . . . . . . . . . . 609
Hela Jebali, Noel Richard, Hermine Chatoux, and Mohamed Naouai

I-HAZE: A Dehazing Benchmark with Real Hazy and Haze-Free

Indoor Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 620
Cosmin Ancuti, Codruta O. Ancuti, Radu Timofte,
and Christophe De Vleeschouwer

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 633

Video Analysis
 Improving a Switched Vector Field Model
for Pedestrian Motion Analysis

Catarina Barata(B) , Jacinto C. Nascimento, and Jorge S. Marques

Instituto de Sistemas e Robótica, Instituto Superior Técnico,

1049-001 Lisboa, Portugal
ana.c.fidalgo.barata@ist.utl.pt

Abstract. Modeling the trajectories of pedestrians is a key task in video

surveillance. However, finding a suitable model to describe the trajecto-
ries is challenging, mainly because several of the models tend to have a
large number of parameters to be estimated. This paper addresses this
issue and provides insights on how to tackle this problem. We model the
trajectories using a mixture of vector fields with probabilistic switch-
ing mechanism that allows to efficiently change the trajectory motion.
Depending on the probabilistic formulation, the motions fields can have
a dense or sparse representation, which we believe influences the per-
formance of the model. Moreover, the model has a large set of param-
eters that need to be estimated using the initialization-dependent EM-
algorithm. To overcome the previous issues, an extensive study of the
parameters estimation is conducted, namely: (i) initialization, and (ii)
priors distribution that controls the sparsity of the solution. The various
models are evaluated in the trajectory prediction task, using a newly pro-
posed method. Experimental results in both synthetic and real examples
provide new insights and valuable information how the parameters play
an important in the proposed framework.

Keywords: Surveillance · Trajectories · Motion fields

Hidden Markov Models · Expectation maximization

1 Introduction

Activity recognition, movement prediction (see Fig. 1), and detection of abnor-
mal behaviors are critical tasks in surveillance applications, requiring appropriate
models to describe them. Since many of the surveillance settings are based on
far-field cameras that do not provide detailed information about the pedestrians
in a video scene, several of these models are based on the characterization of
the trajectories performed by the pedestrians. In fact, trajectories allow us to
identify and collect statistics of typical motions, activities, and interactions in a
video scene [1]. Consequently, having a reliable description of the possible tra-
jectories is of major importance. However, trajectory modeling is a challenging
problem due to its great spatial and temporal variability.
c Springer Nature Switzerland AG 2018
J. Blanc-Talon et al. (Eds.): ACIVS 2018, LNCS 11182, pp. 3–13, 2018.
https://doi.org/10.1007/978-3-030-01449-0_1
4 C. Barata et al.

Various methods have been proposed to efficiently describe trajectories.

Amongst these methods, we are interested in the ones that resort to a genera-
tive approach, i.e, the trajectories are assumed to follow a dynamical equation
and are governed by known motion patterns. In [2–4] a Gaussian process (GP)
regression is used for trajectory modeling under a probabilistic viewpoint. More
specifically, in [3] the authors present an unsupervised algorithm in which GP
is used to estimate the instantaneous velocity of the pedestrian, while in [2] GP
is used to model flow functions. The latter representation also allows for incre-
mentally predicting possible paths and detecting anomalous events from online
trajectories. In [4] it is proposed an incremental and unsupervised approach,
where the model is updated by receiving new trajectory samples. Dirichlet pro-
cesses have also been used to model trajectories [6–8], as well as vector fields
estimated using the k-means algorithm [5]. The latter work demonstrated the
ability of vector fields to describe other types of trajectories besides pedestrian
ones (e.g., hurricanes and cell-phone GPS data).

Fig. 1. Pedestrian trajectory: past (yellow) and possible future motions (green). (Color
figure online)

Our work is related to [9] where vector fields are √ used √ to model pedestrian
trajectories. This method assumes that a grid with n × n nodes is defined in
the image domain. Each node is characterized by the following parameters: (i) a
set of motion vector fields (velocity vectors), (ii) a field of isotropic covariance
matrices, and (iii) a switching probability matrix that governs the transition
between fields in that node. Only one motion field is active at each trajectory
point and the transition between them depends on the spatial location of the
pedestrian in the grid. This method has been show to efficiently characterize tra-
jectories in the task of activity recognition [9]. However, it has some limitations:
(i) the parameter estimation is performed using the EM algorithm that strongly
depends on the initialization, and may encourage wrong estimates; and (ii) the
fields were assumed to be dense and estimated in nodes where information was
not available. Recently, a new formulation was proposed to deal with the lat-
ter issue: imposing a sparsity constraint to the vector fields [10]. However, this
work failed to demonstrate the impact of such strategy at describing pedestrian
trajectories. Moreover, to the best of our knowledge, none of the vector fields
 Improving a Switched Vector Field Model 5

models has ever been used to predict future positions in a trajectory. We believe
that this is a critical task in surveillance applications.
This paper provides valuable and comprehensive contributions, namely: (i)
provides an effective way on how to initialize the vector fields in the EM algo-
rithm, (ii) evaluates different vector fields formulations, and (iii) provides a
new methodology for trajectory prediction, using the space-varying dynamical
model. For the first item, we perform a systematic study comprising the follow-
ing three strategies: (a) random, (b) uniform, and (c) clustering initialization,
thus not jeopardizing deficient estimates related to poor initializations. For the
second item, we compare the vector fields formulations of [9,10]. The proposed
prediction strategy is used to evaluate the initialization and type of fields.

2 Switching Dynamical Model

In this work, the trajectory of a pedestrian is assumed to be represented as a
sequence of positions x = (x1 , . . . , xL ), with xt ∈ [0, 1]2 . We also consider that
all of the possible motions in a scene may be described using K motion fields
Tk : [0, 1]2 → R2 , where k ∈ {1, ..., K} is the identifier of the field. Only one
of these fields is active at a specific time instant, but it is possible to switch
between them. Under these assumptions, each position is generated as follows
xt = xt−1 + Tkt (xt−1 ) + wt , (1)
where kt denotes the active motion field at time t and wt ∼ N (0, σk2t I) is a white
random perturbation with zero mean and isotropic covariance.
Switching between active fields depends on the pedestrian position, and it is
modeled as a first order Markov process
P (kt = j|kt−1 = i, xt−1 ) = Bij (xt−1 ), (2)
where B(x) = {Bij (x), i, j ∈ {1, . . . , K}} is a stochastic space-varying matrix of
switching probabilities (transition matrix) that depends on the position x, i.e.,
switching probabilities at different positions are usually different.
We specify the motion fields Tk and the transition
 matrix B using a regular
grid on the image G = gi ∈ [0, 1]2 , i = 1, . . . , n . This means that each node gi
of the grid is associated with K velocity vectors Tki and a transition matrix B i .
Outside of the nodes, the velocity vector Tk (x) and switching matrix B(x) are
obtained by bilinear interpolation, as detailed below
Tk (x) = Tk Φ(x), (3)

n

B(x) = B i φi (x), (4)
i=1

where Tk is the dictionary of node velocities Tki associated with the k − th field,
B i is the switching matrix associated with node gi , Φ(x) is a N × 1 normalized
sparse vector of known interpolation coefficients, and φi (x) is its i − th element.
6 C. Barata et al.

If we assume a L-sequence trajectory and a known sequence of active fields

k = (k1 , ..., kL ), it is easy to compute the joint probability p(x, k)
L

p(x, k|θ) =p(x1 , k1 ) p(xt |kt , xt−1 )p(kt |kt−1 , xt−1 )
t=2

= p(x1 )P (k1 ) N (xt |xt−1 + Tkt (xt−1 ), σk2t I)
× Bkt−1 ,kt (xt−1 ), (5)

where θ = (T, B, σ 2 ) denotes the complete set of model parameters. From (5),
it is clear that the sequence of active fields k = (k1 , ..., kL ) is modeled as a
realization of a first order Markov process, with some initial (known) distribution
P (k1 ) and a space-varying transition matrix, i.e., Bkt−1 ,kt (xt−1 ). In the following
section we address the estimation of the model parameters.

3 MAP Estimation of the Model

Lets assume that we have a set of S observed trajectories, denoted as X =

 (1)  (j) (j)
x , . . . , x(S) , where x(j) = (x1 , ..., xLj ) is the j-th trajectory with Lj sam-
ples. Naturally, the corresponding sequences of active fields K = {k (1) , ..., k(S) }
are missing. Given this set, our goal is to estimate the model parameters
θ = (T, B, σ 2 ) using the MAP formulation
 
θ̂ = arg max log p(X|θ) + log p(θ) , (6)
θ

K
where log p(θ) = k=1 log p(θk ) is the log prior distribution. Assuming indepen-
dence between the various parameters, it is possible to further decompose the
prior as follows
K
 K

log p(θ) = log p(Tk ) + log p(σk2 ) + log p(B). (7)
k=1 k=1

The complete likelihood log p(X, K|θ) may also be easily defined from (5),
by computing a sum over all the trajectories. However, its marginalizing over all
the admissible labels K cannot be computed.
The aforementioned limitation may be addressed using the Expectation-
Maximization (EM) method to estimate the model parameters. This amounts
to iteratively maximizing the auxiliary function

U (θ, θ ) = E log p(X, K|θ)|K, θ(t) + log p(θ) (8)

with respect to θ and taking into account the available estimates θ from the
previous iteration.
 Improving a Switched Vector Field Model 7

Using (5) and replacing into (8) we obtain the following expression (disre-
garding the constants)

 L 
S  K
U (θ, θ ) = −
(s)
wk (t) log N (xt |xt−1 + Tkt (xt−1 ), σk2t I)
s=1 t=2 k=1
S 
 L 
K
+ wp,q
(s)
(t) log Bkt−1 ,kt (xt−1 )
s=1 t=2 p,q=1
K
 K

+ log p(Tk ) + log p(σk ) + log p(B), (9)
k=1 k=1

where wp (t) = P (kt = p|x(s) , θ ) and wp,q (t) = P (kt−1 = p, kt = q||x(s) , θ ).

(s) (s) (s) (s) (s)

These weights are computed in the E-step, using the forward-backward algorithm
proposed in [12]. The optimization with respect to θ is performed in the M-step,
and involves a separate maximization with respect to T, B, and σ 2 [9].
As discussed in [10] the prior log p(θ) may be used to incorporate a priori
knowledge about the parameters, as well as to enforce specific properties. In
particular, efforts have been made to define a prior distribution p(T) that could
lead to smooth and sparse fields:

log p(Tk ) = αΔTk 22 + βTk pp , (10)

where Δ is an operator that computes all differences between velocities of neigh-

boring nodes and .p denotes the pth norm. Here, the first term sets that any
neighbor grid nodes xg1 , xg2 ∈ G should have similar estimated velocities, i.e.,
the difference Tkt (xg1 ) − Tkt (xg2 ) should be small. The second term enforces
small velocity values in most of the nodes. These values may be zero if p = 1,
leading to sparse vector fields where the velocities are only estimated for nodes
that are supported by observations [10]. The values chosen for the regularization
constants α and β and norm p define the importance of each term, and may
influence the estimation of the fields Tk . However, the assessment of their role
and relevance has never been performed.
Besides the priors, another criteria that strongly influences the estimation of
the model is the initialization of the parameters θinit , since the EM algorithm
is very sensitive to initialization [11]. In the following section we propose an
experimental framework to evaluate the role of these criteria in the estimation
of the motion models.

4 Experimental Setup

The goal of this work is to provide an extensive experimental study on the role of
both initialization and the prior p(T) in the estimation of the motion models. To
conduct such a study it is necessary to (i) identify the types of initialization and
priors to be tested; (ii) select one or more datasets to conduct the experiments;
8 C. Barata et al.

1 1

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

Fig. 2. Experimental datasets: synthetic examples (1st row) and real pedestrian tra-
jectories (2nd row).

and (iii) define an evaluation metric. In this section we address each of these
requirements.
(i) Initialization: The EM-algorithm, which is used in the estimation of
the parameters, strongly depends on initialization. Thus, we will evaluate three
types of initialization: (i) random, where each node of a field is initialized with
a random velocity vector; (ii) uniform, where each node of a field is initialized
with the same (random) velocity vector; and (iii) clustered, where we perform a
preliminary clustering of the velocity values and set the number of centroids to
be equal to the number of fields K.
(ii) Prior: Another goal of this paper is to provide a comprehensive study
of the role of the fields’ prior (10). This means that for each of the aforemen-
tioned initializations, we train several models using different combinations of
α ∈ {0.5, 1, 2}, β ∈ {0, 0.05, 0.1, 0.15}, and p ∈ {1, 2}, leading to a total of 21
models per initialization type.
(iii) Datasets: Our study is conducted using four datasets (two synthetic
and two real). The first synthetic case comprises K = 3 motion types and
100 randomly generated trajectories, as is exemplified in Fig. 2 (1st row-left).
The trajectories were generated according to the following procedure. First, the
starting point was randomly chosen between a small region around [0, 0.9]T and
[0.5, 0]T . Conditioned on the selected point, the motion regime can either be a
down motion with a slope of 45◦ or an up motion. When the trajectories reach
 Improving a Switched Vector Field Model 9

the center of the image there is a probability of switching between these motions
and a third one, oriented form left to right.

1
L
Table 1. Dataset description. L̄ is the average trajectory length and v̄ = L t=1 xt −
xt−1 22 is the average displacement.

Dataset # Trajectories L̄ v̄ (×10−3 )

Synthetic 1 100 27 ± 5 46 ± 4
Synthetic 2 100 77 ± 40 21 ± 0.4
Stairway 86 57 ± 18 12 ± 2
Campus II 1083 50 ± 48 2 ± 0.9

The second synthetic example mimics a roundabout with four entrance/exist

points, as exemplified in Fig. 2 (1st row-right). Three motion regimes (K = 3) are
considered: a circular and counterclockwise one, and two linear movements that
represent, respectively, the entrance and exiting of the roundabout. Switching
may occur at any of the entrance/exit points. As before, we randomly generate
100 trajectories.
The two real datasets were acquired at the IST university campus in Lisbon
(see Fig. 2 (2nd row)), using static far field cameras. The first dataset (2nd row-
left), which we call Stairway, comprises 86 trajectories of pedestrians going up
and down a stairway, while the second dataset, called Campus II (2nd row-right),
comprises more than 900 trajectories of pedestrians crossing the IST campus.
Two cameras were used to acquired the Stairway dataset, while the Campus II
dataset was acquired using only one. In the first case, two homographies were
performed to i) align the images and (ii) correct the distortions cause by the
perspective projection. Regarding the Campus II dataset, it was only necessary
to correct the distortions. We set the number of motion fields to be estimated to
K = 2 for the stairway example (up and down), and to K = 4 in the Campus
II set (roughly north, south, east, and west).
In Table 1 we summarize the properties of each dataset, namely the number
of trajectories, the mean trajectory length (L̄), and the average displacement
between two consecutive trajectory points.
(iv) Evaluation - prediction: Finding an appropriate metric to evaluate the
role of the initialization and prior is a challenging task. One could assess the error
in the estimation of the model parameters, but such metric is suitable only when
we know the ground truth values that generated the trajectories. Such information
is only available in the case of the synthetic data. Alternatively, we can assess the
ability of the estimated models to predict the future of the trajectory, i.e., given a
sequence of past positions (x(1), ..., x(t0 )) we want to predict the position x̂(t0 +δ),
δ steps ahead in time using the dynamical model (1).
In our case, this means that we have to predict not only the position, but also
the most probable motion field k ∗ that leads to it (see Fig. 1). To achieve this
10 C. Barata et al.

goal, we apply the forward-backward algorithm [12] used in the E-step of model
estimation, namely the forward part, that works as follows. First, lets consider
the past, i.e., t ≤ t0 , and assume the existence of a set of forward variables

αi (t) = P r{x(1), ..., x(t), kt = i|θ̂}, (11)

such that α(t) = [α1 (t), ..., αK (t)] is a vector that stores all of the forward
probabilities of time instant t. For a given past instant, α(t) is estimated as
follows:

α(t) = D(xt )B(xt−1 )T α(t − 1), t = 2, ..., t0 , α(1) = D1 π, (12)

D(xt ) ∈ RK×K is a diagonal matrix with

Dii = N xt ; xt−1 + Ti (xt−1 ), σi2 I , i = 1, ..., K. (13)

For t > t0 the real position x(t) is unknown. Therefore, matrix D can not be
computed and the estimation of the forward probabilities α̂(t) is given by:

α̂(t) = B(x̂t−1 )T α̂(t − 1), t = t0 + 1, ..., t0 + δ, (14)

where
x̂t = x̂t−1 + Tkt∗ (x̂t−1 ) + 0. (15)
kt∗ denotes the active field at each instant t > t0 , and is defined as: k ∗ =
maxk αk (t).
We use the displacement error (DE) as the evaluation metric:
L−δ

DE = x(t0 + δ) − x̂(t0 + δ)22 , (16)
t0 =1

where x̂(t0 + δ) is computed by recursively applying (14) to estimate the best

field and (15) to predict the position. We set δ = 5, as proposed in other works
(e.g., [13]).

5 Results

In Fig. 3 we show the best performances for each dataset (synthetics 1 and
2, Stairway, and Campus II). These results were obtained using leave-one-
trajectory-out cross validation and the performances were computed for each
pair (initialization,prior ). Here dense means that β = 0, while l2 and l1 corre-
spond to p = 2, 1, in (10). Table 2 shows the best overall performances, as well
as the combination of model parameters and initialization that led to it. These
results demonstrate that the proposed prediction strategy is able to successfully
the estimate the position of a pedestrian δ = 5 time steps ahead, achieving a
low DE in all datasets.
 Improving a Switched Vector Field Model 11

0.014
Syn. 1 + dense
0.012 Syn.1 + l2
Syn. 1 + l1
0.01 Syn. 2 + dense
Syn. 2 + l2
0.008 Syn. 2 + l1
DE

Stairs. + dense
0.006 Stairs. + l2
Stairs + l1
0.004 Campus + dense
Campus + l2
0.002 Campus + l1

0
Random Uniform Clustering
Initialization

Fig. 3. Experimental results for synthetic examples (red and blue) and real pedestrian
trajectories (green and magenta). (Color figure online)

Table 2. Best experimental results out of 63 possible configurations for each dataset.

Dataset DE Prior Best Configuration

Synthetic 1 5.0 ± 11.0 × 10−3 Dense Random, α = 0.5, β = 0
4.7 ± 10.9 × 10−3 l2 Random, α = 0.5, β = 0.1
4.0 ± 9.0 × 10−3 l1 Random, α = 0.5, β = 0.05
Synthetic 2 1.1 ± 0.80 × 10−2 Dense Random, α = 1, β = 0
7.4 ± 5.8 × 10−3 l2 Random, α = 0.5, β = 0.15
4.6 ± 3.0 × 10−3 l1 Random, α = 0.5, β = 0.15
Stairway 3.7 ± 5.8 × 10−3 Dense Clustering, α = 2, β = 0
3.7 ± 6.4 × 10−3 l2 Clustering, α = 0.5, β = 0.05
4.4 ± 8.4 × 10−3 l1 Clustering, α = 1, β = 0.05
Campus II 2.2 ± 0.5 × 10−4 Dense Clustering, α = 2, β = 0
2.4 ± 0.6 × 10−4 l2 Clustering, α = 0.5, β = 0.15
2.4 ± 0.5 × 10−4 l1 Random, α = 0.5, β = 0.05

A more thorough analysis of the results suggest that the initialization method
has a significant impact in the synthetic and Stairway datasets, while the per-
formance in the Campus II remains almost unchangeable. The role of the ini-
tialization is particularly notorious for synthetic example 2 (the roundabout),
where velocity clustering leads to a large increase in the DE. This higher error
explained by the existence of a circular movement that is inefficiently modeled
by velocity clustering.
Enforcing sparsity in the estimation of the fields (p = 1) leads to improved
prediction results for the synthetic data. However, the opposite is observed for
the real datasets, where dense fields achieve the best results. Such disparity may
be due to the proportionally higher variability that exists in pedestrian velocities,
when compared with the synthetic datasets (see Table 1). Setting p = 2 in (10)
12 C. Barata et al.

has some impact in the performance, but it is clearly outperformed by the other
priors. Regarding the prior regularization constants, α and β, it seems that
setting α = 0.5 and β = 0.05 is the preferred combination for the sparse prior
in three out of the four datasets. In the case of the dense prior, we observe that
setting α to a higher value leads to the best results for the real datasets, while
no trend is observed for the synthetic ones.

6 Conclusions
Efficiently modeling the trajectories of pedestrians is critical for most surveil-
lance setups. However, several of the models proposed to tackle this issue rely
on complex methods that require the tunning of multiple parameters. In this
paper we addressed this issue, taking as starting point a motion model based on
switching motion fields.
Our contributions are two-folded. First we proposed a trajectory prediction
strategy for the aforementioned method, which allows a qualitative evaluation of
the method. Second, we conducted a thorough study of the model parameters,
namely initialization, velocity priors, and constants, in order to obtain insights
of their role. Both contributions are original of this paper. Experiments were
conducted in four datasets and the obtained results are relevant and promising.
Future work should rely on exploring the applicability of the motion fields to
other tasks, such as the detection of abnormal trajectories or activities.

Acknowledgments. This work was supported by the FCT project and plurianual
funding: [PTDC/EE- IPRO/0426/2014], [UID/EEA/50009/2013].

References
1. Aggarwal, J., Ryoo, M.: Human activity analysis: a review. ACM Comput. Surv.
43(3), 16 (2011)
2. Kim, K., Lee, D., Essa, I.: Gaussian process regression flow for analysis of motion
trajectories. In: Proceedings of IEEE International Conference Computer Vision
(ICCV) (2011)
3. Ellis, D., Sommerlade, E., Reid, I.: Modelling pedestrian trajectory patterns with
Gaussian processes. In: Proceedings IEEE International Conference Computer
Vision (ICCV) Workshops (2009)
4. Bastani, V., Marcenaro, L., Regazzoni, C.S.: Online nonparametric bayesian activ-
ity mining and analysis from surveillance video. IEEE Trans. Image Process. 25(5),
2089–2102 (2016)
5. Ferreira, N., Klosowski, J.T., Scheidegger, C.E., Silva, C.T.: Vector field k-means:
clustering trajectories by fitting multiple vector fields. In: Eurographics Conference
on Visualization, vol. 32(3), pp. 201–210 (2013)
6. Bastani, V., Marcenaro, L., Regazzoni, C.: Unsupervised trajectory pattern clas-
sification using hierarchical Dirichlet process mixture hidden Markov model. In:
Proceedings of IEEE International Workshop Machine Learning for Signal Pro-
cessing (MLSP), p. 16 (2014)
 Improving a Switched Vector Field Model 13

7. Hu, W., Li, X., Tian, G., Maybank, S., Zhang, Z.: An incremental DPMM-based
method for trajectory clustering, modeling, and retrieval. IEEE Trans. Pattern
Anal. Mach. Intell. 35(5), 10511065 (2013)
8. Lin, D.: Online learning of nonparametric mixture models via sequential variational
approximation. In: Advances in Neural Information Processing Systems, pp. 395–
403 (2013)
9. Nascimento, J.C., Figueiredo, M.A.T., Marques, J.S.: Activity recognition using
mixture of vector fields. IEEE Trans. Image Process. 22(5), 1712–1725 (2013)
10. Barata, C., Nascimento, J.C., Marques, J.S.: A sparse approach to pedestrian
trajectory modeling using multiple motion fields. In: International Conference on
Image Processing (2017)
11. Figueiredo, M.A.T., Jain, A.K.: Unsupervised learning of finite mixture models.
IEEE Trans. Pattern Anal. Mach. Intell. 24(3), 381–396 (2002)
12. Rabiner, L.R.: A tutorial on Hidden Markov Models and selected applications in
speech recognition. Proc. IEEE 77(2), 257–286 (1989)
13. Alahi, A., et al.: Learning to predict human behaviour in crowded scenes. In: Group
and Crowd Behavior for Computer Vision, pp. 183–207 (2017)
Another random document with
no related content on Scribd:
Amblyoponides, 132, 180
Ambrosia, 295
Ambryllis, 409
Amicta quadrangularis, 394
Ammophila, 111;
A. affinis, 111;
A. hirsuta, 111;
A. holosericea, 111
Amphicyrtides, 242
Amphidasis betularia, 412, 414
Amphiodont, 193
Amphipneustic, 450
Amphizoa lecontei, 207
Amphizoidae, 207
Ampulex compressa, 114, 115;
A. ruficornis, 115;
A. sibirica, 114
Ampulicides, 114 f., 169
Amycterides, 291
Anal armature, 328, 416
Anal nervures, 318
Anaphe, 376
Anaspini, 267
Anaspis, 268
Anatomy—see External Structure and Internal Anatomy
Anchor-process, 459
Ancient, Lepidopteron, 435—see also Primitive
Ancylolominae, 425
Andrena, 23, 25, 30, 301, 303;
hair of, 11;
A. labialis, 488;
A. nigroaenea, 23;
A. ovina, 30
Andrenidae, 20
Andrenides, 23
Andrenimorpha, 388
Andrenoides, 20
Androconia, 331 f.
Anergates atratulus, 160 f.
Angelitos, 63
Anisopteryx aescularia, 411
Anisotoma cinnamomea, 222
Anisotomidae, 223
Anisotomides, 222
Anlagen, 143
Anobiides, 247
Anobium, 254;
A. paniceum, 247;
A. striatum, 248;
A. tessellatum, 248
Anochetus ghiliani, 174
Anomma, 178
Anophthalmus, 205
Anoplura, 599 f.
Anosia erippus, A. menippe, 345;
A. plexippus, 307;
larva, 324;
pupa, 327
Antennae;
of Belostoma, 566;
of flies, 441;
of Lamellicornia, 191;
of Lepidoptera, 307;
of butterflies, 340, 341;
of Sphingidae, 380
Antherophagus, 235
Anthicidae, 266
Anthidium, 45;
A. bellicosum, 47;
A. diadema, 45;
A. manicatum, 45;
A. septemdentatum, 47;
A. strigatum, 30
Anthocopa, 51
Anthocoridae, 560
Anthomyia angustifrons, A. brassicae, A. cana, 506
Anthomyiidae, 506
Anthophila, 10 f.
Anthophora, 32, 33;
destroyer of, 272, 274;
A. personata, 33;
A. pilipes, 33;
proboscis of, 17
Anthothrips aculeata, 530
Anthracides, 486
Anthrax, 486 f.;
A. fenestralis, 489;
A. trifasciata, 44
Anthrenus fasciatus, 241
Anthribidae, 278, 290
Antisquama, 448
Antitegula, 447
Ant-plant, 138, 139, 168
Ants, 131 f.;
and Aphidae, 590;
and caterpillars, 356
Ants'-nest, Insects, 200, 213, 221, 223, 224, 225, 231, 236, 240,
548;
larva, 501, 502
Anus, 314, 320
Aorta, 320
Aortal chamber, 320
Apate capucina, 246
Apatela, 418
Apathus = Psithyrus, q.v.
Apatidae, 246
Apatura;
larvae, 354;
A. iris, 344
Apaturides, 352
Aphaenogaster, 164, 165, 221;
 A. arenarius, 164;
A. barbarus, 131, 164;
A. structor 164, 240
Aphaniptera, 522 f.
Aphanocephalus, 228
Aphelocheirus, 565
Aphidae, 581 f.;
and ants, 181
Aphidiides, 590
Aphis maidi-radicis, 584
Aphomia sociella, 424
Apidae, 10 f., 20, 32
Apioceridae, 492
Apis, 53;
A. adansonii, 69;
A. domestica, 68;
A. dorsata, 69;
A. fasciata, A. ligustica, 68;
A. mellifica, 65 f.;
feet of queens and workers, 69;
ligula of, 16;
worker and hairs, 12
Apoda, 402;
A. testudo, etc., 401, 402
Apodidae, 402
Apoica pallida, nest, 83
Aporia crataegi, 322
Appetite, 491
Apterogyna, 96
Apterona, 393, 394;
A. crenulella, var. helix, 395
Apterous, beetles, 187, 263;
females, 95, 96, 140, 171, 174, 315, 392, 393, 407, 413, 430,
592—see also Workers;
Insects, 95, 96, 474, 495, 496, 505, 518, 531, 581;
males, 140, 160, 161, 172, 585, 594
Aquatic;
 caterpillar, 377;
cocoons, 280;
larva, 421 f., 425, 504;
pupa, 423
Aradidae, 550
Aradus orientalis, 550
Araeocerus, 290
Araschnia levana, A. prorsa, 353
Arbelidae, 369, 396
Arceina, 373
Archiapidae, 21, 22
Archiapides, 21 f.
Arctia caja, 308;
A. villica, 410
Arctiidae, 370, 404, 408, 410
Argiva, 414
Argynnis, larvae, 354;
A. paphia, androconia, 332
Argyromoeba sinuata, 76;
A. trifasciata, 486
Arista, 442
Army-worm, 416
Aromia moschata, 188
Arthropterus, 214
Asclera caerulea, 267
Ascodipteron, 520
Asemorhoptrum lippulum, 160
Asilidae, 491
Asilus, 492;
A. crabroniformis, 441
Asparagus-beetle, 281
Aspidiotus camelliae, 592;
A. nerii, 595, development of, 596
Aspidiphorus, 246
Aspidomorpha, pupa, 283, 284
Association, of Anergates and Tetramorium, 160;
of ants and other Insects etc., 180 f.;
 of Formica and Formicoxenus, 159;
of Strongylognathus and Tetramorium, 162;
of Tomognathus and Leptothorax, 161
Astata boops, 119
Astatides, 119
Asteidae, 504
Astomella lindeni, 490
Astynomus, 285
Atemeles, 225;
and ant, 182
Athericerous, 441
Atherix, 481;
A. ibis, 480
Athous rhombeus, 257
Atractocerus, 254
Atrophy of mouth and stomach, 310
Atta, 137, 164 n., 165, 502
Attacus, 373;
A. atlas, 373
Attelabides, 291
Attini, 158, 159, 165
Attitude, 381, 384, 385, 388, 412, 413, 425, 429
Atylotus fulvus, 483
Autocrates aenea, 275
Automeris, 373
Axin, 598
Azteca, 158
Azygos oviduct, 321

Badamia exclamationis, 365

Barrett, on increase of melanism, 414
Basket-worms, 393
Bat-parasites, 521, 560
Bataillon, on metamorphosis, 306
Bates, H. W., on classification of butterflies, 344;
on homoeochromatism, 351;
on Megacephala, 201
Batesian mimicry, 337, 339
Bathyscia, 221
Beak, 532
Beauregard, on vesicating Insects, 275
Beaver, Insect on, 219, 221
Becher, on mouth of Diptera, 444 n.
Bed-bug, 559;
enemy of, 558
Beddard, on animal coloration, 339 n.
Bee (i.e. honey-bee)—see Apis mellifica
Bee-louse or -tick, 520
Bees, 10 f.
Bees born of carcases, myth, 499
Bees' nest beetle, 235
Bees, stylopised, 300, 303
Beetles, 184 f.
Bellesme, on buzzing, 19
Belostomidae, 534, 565
Bembecidae, 482
Bembecides, 119 f.
Bembex, 509;
B. rostrata, 4, 120, 120 f.;
B. spinolae, 130 n.
Benchucha bug, 559
Berosus, 218
Berytidae, 548
Bibio, 475, 476, 477;
B. marci, 477
Bibionidae, 475
Birds and butterflies, 338
Biscuit-weevil, 247
Bitoma crenata, 233
Bittacomorpha, 473
Black-fly, 530
Blanchard, on flies attacking man, 517 n.
Blepharoceridae, 464
Blind beetles, 205, 221, 233
Blissus leucopterus, 548
Blister-beetles, 269
Blochmann, on founding new nests, 145
Blood-sucking, Diptera, 457;
Mosquitoes, 467
Blood-worms, 468
Blow-flies, 511
Blue-bottles, 511
Bogus Yucca-moth, 433
Boletophila luminosa, 463
Boll-worm, 416
Bombardier-beetles, 201
Bombus, 53 f.;
insect in nest of, 221;
fleas in nests of, 525;
parasite of, 94, 497;
proboscis of, 13 f., 14;
B. agrorum, 54;
B. lapidarius, 54;
B. muscorum, 57;
B. variabilis, 60
Bombyces, 367
Bombycidae, 368, 375, 406
Bombyliidae, 485
Bombylius major, 488
Bombyx mori, 375;
B. yamamai, 325
Book-worm, 247
Borboridae, 504
Borborus, 505
Borocera madagascariensis, 405
Bostrichidae, 246
Bot-flies, 514
Brachelytra, 224
Brachycera, 441, 454
Brachycerides, 291
Brachyscelides, 592, 598
Brachytarsus, 290
Braconidae, 590
Bradypus cuculliger, Tineid on, 430
Brahmaeidae, 368, 374
Brain, 320;
cephalic and thoracic, 449
Branchiae, 208, 244
Brands, 332
Brassolides, 349
Brathinus, 223
Brauer, on Dipterous larvae, 451;
on Oestridae, 514
Braula coeca, 520
Braulidae, 520
Breastbone, 459
Breeze-flies, 443, 481
Breitenbach, on proboscis of Lepidoptera, 311 n.
Breithaupt, on proboscis of bee, 15;
on deglutition of bees, 18
Brenthidae, 295
Brenthus anchorago, 297
Brephos notha, 415, 416
Brimstones, 357
Brontes planatus, 234
Brown-tail moths, 407
Bruchidae, 276
Bruchus fabae and B. lentis, 277, B. pisi, 277
Bryophila, 418
Buckell, on development of pattern, 335
Buffalo-gnats, 477
Bugong-moth, 417
Bugonia-myth, 499
Bull-dog ants, 171, 173
Bull's-horn thorn and ants, 168
Bumble bee—see Bombus
Buprestidae, 261
Buprestis attenuate, supposed larvae of, 262 n.
Burgess, on suction, 311
Burnet-moths, 390
Burrows, of Dasypoda, 27;
of Halictus, 24, 25;
of Odynerus, 74
Bursa copulatrix, 321
Burying-beetles, 221
Butterflies, 341 f.
Buzzing, 19
Byrrhidae, 242, 255
Byrrhus pilula, 242
Bythoscopidae, 578
Byturus, 241

Cadphises moorei, 391

Calandrides, 289
Calcium oxalate, 406
Calicurgus, 101;
C. hyalinatus, 102, 106
Caligo eurylochus, 350
Callidea baro, 303 n.
Callidulidae, 370, 400
Calliphora; 448;
C. erythrocephala, C. vomitoria, 511
Callirhipis dejeani, 256
Callomyia, 496
Callostoma fascipennis, 489
Caloptenus, 270;
C. italicus, 489, C. spretus, 488, 506, enemies of
Calypter, calypterate, 448
Calyptrate Muscidae, 448, 504
Camberwell Beauty, 352
Camel bot-fly, 515
Camponotides, 144
Camponotus, 145;
C. ligniperdus, 138, 145, 147;
C. pennsylvanicus, 138, 146;
 C. rubripes, 131;
C. rufipes, 137
Camptosomes, 279, 281
Canephorinae, 394, 395
Cantharidae, 269 f.
Cantharides, 270
Capsidae, 561
Capsus laniarius, 539
Carabidae, 204 f.
Carabides, 206
Caraboidea, 190, 200 f.
Carcinocoris, 554
Carder-bees, 45 f., 45
Cardiocondyla, 161
Cardiophorus, 258
Carlet, on sting, 6;
on sound-organs of Cicada, 574
Carnivora, 200
Carotine, 549
Carpenter-bees, 33
Carpenter-worms, 395
Carpets, 411
Carpocapsa juliana, C. pomonella, C. splendana, C. saltitans,
428
Carpophagus, 278
Carteria lacca, 597
Carus, on paedogenesis, 461
Caryoborus, 278
Case, 281, 392, 393, 394, 417, 422, 423, 430, 431
Cassidides, 279, 283
Caste-production, 142
Castnia, 307, 309, 316, 319;
C. eudesmia, C. therapon, 372
Castniidae, 369, 371
Castor canadensis, parasite of, 219
Cataclysta lemnata, 423
Caterpillar, 322, 324, 325;
 of Diptera, 474
Catopomorphus, 221
Cauda, 538, 588
Cave-beetles, 205, 221
Cebrionides, 260
Cecidipta excoecaria, 424
Cecidomyia buxi, 459;
C. destructor, 460;
C. tritici, 460
Cecidomyiidae, 455 n., 458
Cecropia, plant and ants, 158
Cedeocera, 297 n.
Cell, of wing, 317, 318;
complete and incomplete, 116 n.
Cells, formation of, by bees, 21, 22, 24, 25, 28, 33, 34, 35, 46,
48, 51, 52, 54, 56, 60;
earthen, 72, 106;
of Coelonites, 89
Celyphidae, 504
Celyphus, 505
Cemonus unicolor, 128
Cephaloidae, 275
Cephalomyia maculata, 515
Cephaloon, 275
Cephalothorax, 465;
of Stylops-larva, 302
Cephenomyia rufibarbis, 517
Cerambycidae, 278, 285
Cerambycides, 287
Ceramius, 89;
C. lusitanicus, 89
Ceranchia, 374
Cerapachys, 175 n.
Ceratina, 11, 32
Ceratocampidae, 368, 375
Ceratocombidae, 559
Ceratognathini, 194, 195
Ceratonema, 401
Ceratopogon, 469;
C. bipunctatus, C. pulicaris, C. varius, 470
Cerceris, 125;
C. arenaria, 125;
C. bupresticida, 125;
C. labiata, 125;
C. tuberculata, 126
Cercopidae, 577
Ceresa bubalus, C. taurina, 577
Cerocoma schaefferi, 275
Cerophytides, 260
Ceroplastes ceriferus, 597
Ceroplatus mastersi, 463
Cerura vinula, 383
Cervical sclerites, 472
Cetonia, in ants'-nests, 149;
C. floricola, 200
Cetoniides, 195, 199
Chaerocampa, 380; recte Choerocampa
Chaetophorous, 446
Chaetotaxy, 446
Chafers, 194 f.
Chalcosiidae, 391, 420
Chalia hockingi, 394
Chalicodoma, 32, 35;
C. muraria, 30, 35 f., 36, 254, 486;
C. parietina, C. pyrenaica, 39
Change of habit in larva, 301, 431
Chapman, Dr. T. A., on Chrysis, 3;
on classification of pupae of Lepidoptera, 367;
on Hepialus, 398;
on Metoecus paradoxus, 268;
on pupa of Lepidoptera, 327 n.
Charagia, 396
Chartergus chartarius, nest of, 82, 83
Cheilosia chrysocoma, 439
Cheimatobia brumata, 414
Cheliomyrmex, 180
Chelonariides, 242
Chelonia, 410
Chelostoma, 35
Chennium bituberculatum, 224
Chermes, 583, 586, 587;
C. abietis, 586, 587, 589
Cheshire, on proboscis of bee, 15
Chigger, 525
Child, on sense-organ, 442
China-marks, 421
Chinch-bug, 548
Chionea araneoides, 474
Chiromyzidae, 479
Chironomidae, 468, 474
Chironomus, 440, 468
Chlamydes, 279
Chlorion, 110ü
Chloropidae, 504
Chlorops, 504
Choerocampa elpenor, 380
Choerocampini, 381
Cholodkovsky on Chermes, 586
Choragus sheppardi, 290
Chorion, 322
Chrysalis, 326, 344
Chrysauginae, 423
Chrysididae, 1 f.
Chrysiridia madagascariensis, 419
Chrysis bidentata, 3;
C. ignita, 3;
C. shanghaiensis, 4
Chrysochus pretiosus, 279
Chrysocoris grandis, 303 n.
Chrysomelidae, 276, 278 f.
Chrysomelides, 279
Chrysopolomidae, 396
Chrysops, 482
Cicada, 123;
C. plebeia, 574;
C. septendecim, 569
Cicadellinae, 578
Cicadidae, 568 f.
Cicindela hybrida, 202
Cicindelidae, 201 f.
Cicinnus, 378 n.
Cilix glaucata = spinula, 401
Cimex, 560;
C. lectularius, 559
Cimicidae, 559
Cioidae, 245
Cis melliei, 245
Cistelidae, 264
Cistus salvifolius, beetle-larvae in, 282
Cithaerias, 348
Citheronia, 375
Cixiides, 576
Cixius, 575
Clambidae, 223
Clasper, 314
Classification, of ants, 144;
of bees, 20;
of butterflies, 341;
of Coleoptera, 189;
of Diptera, 454 f.;
of Hemiptera, 543;
of Hymenoptera Aculeata, 10;
of Lepidoptera, 339 f.;
of moths, 366 f.
Clavicornia, 189, 213, 265
Claviger testaceus, 224
Clavigerides, 224
Clavus, 539
Clear-wings, 386
Cleggs, 481
Cleosiris, 400
Cleptes, 2, 4
Cleridae, 253
Click-beetles, 256
Clicking butterfly, 354
Clidicus, 223
Clisiocampa neustria, 322
Clothes-moths, 430
Clouded-yellows, 357
Clypeus, 307
Clythra in ants'-nests, 149
Clythrides, 279
Cnemidotus caesus, 209
Cnethocampa processionea, 376
Coarctate larva, 271
Coccidae, 592 f.;
destroyer of, 290
Coccidula, 239
Coccinellidae, 237
Coccus cacti, 598;
C. mannifera, 597
Cochineal Insect, 598
Cochliopodidae, 402 n.
Cochlophora, 394
Cockchafer, 198
Cockroach, parasite of, 269
Cock-tail, 225
Cocoon, 46, 55, 66, 109, 122, 328, 347, 373, 376, 384, 385,
391, 403, 404, 405, 407, 419, 424, 436, 460, 462, 494;
flax-seed, 460;
of ants, 134;
aquatic, 280, 377
Cocytia durvillii, 382
Cocytiidae, 382
Codling-moth, 428
Coelioxys, 31
Coelonites, 89;
C. abbreviatus, cells of, 89
Coenomyia ferruginea, 480
Coenomyiidae, 479
Coenonympha, 348
Colaenis, 351
Coleophora, 431
Coleoptera, 184 f.
Colletes, 22;
C. daviesanus, 30
Colobopsis, 138
Colon, 320
Colorado beetle, 278
Colour, corresponding with locale, 201;
and surroundings, 337;
of larva and habits, 336;
of Sphingidae larvae, 381;
physiology of larval, 413;
of caterpillars and sex, 325;
development of, in Hemiptera, 542;
of eyes, 440
Coluocera formicaria, 240
Colydiidae, 233, 234
Colydium, 233
Comb, 63, 64, 65, 78, 79
Combs and brushes, 134
Compound pupa, 452
Composmyia, 512
Comstock, on nervures, 317 n.
Conchylidae, 427
Connexivum, 538
Conopidae, 497, 504
Conorhinus sanguisuga, 559
Copiopteryx, 373
Copium clavicorne, 550
Copius, 547
Copper butterflies, 356
Coprides, 195 f.
Coprini, 195 n.
Copris hispanus, 197
Copulatory pouch, 320, 321
Cordyluridae, 504
Coreidae, 546
Corethra, 467
Corium, 539
Corixa, 567
Corixidae, 567
Corn-leaves, larva on, 281
Coronidia, 419
Corotoca, 227
Corticaria, 240
Corylophidae, 228
Corynetides, 253
Coscinocera hercules, 372
Cossidae, 369, 395
Cossonides, 294
Cossus, 309;
C. ligniperda, 319
Costal nervure, 318
Cotton-stainer bug, 548
Cotton-worm, 416
Courtship, 494;
of Hepialus, 398 f.
Coxa, 307
Crabro, 129;
C. cephalotes, 129;
C. stirpicola, 130 n.
Crabronides, 128 f.
Crambidae, 425
Crane-flies, 471
Cratoparis, 290
Cremaster, 327, 328, 344, 426
Cremastochilini, 200
Cremastochilus, 200
Cremastogaster, 165, 213;
C. tricolor, 165
Crepitation, 213, 214
Criocerides, 279, 280
Crioceris asparagi, 281;
C. merdigera, 281
Crossocerus, 130;
C. wesmaeli, 130
Cryptocephalides, 279
Cryptocephalus, 282
Cryptocerata, 544, 562 f.
Cryptocerini, 132, 134, 158, 159, 169
Cryptocerus, 138;
C. atratus, 170
Cryptophagidae, 235, 237
Cryptophagus dentatus, 235
Cryptostomes, 279, 282
Cteniza ariana, destroyer of, 490
Ctenophora, 475
Ctenostylidae, 517
Ctenuchinae, 409
Cuckoo-bees, 22
Cuckoo-spit, 577
Cucujidae, 232, 234
Cucujos, 258
Cuculinae, 20
Culex pipiens, 466
Culicidae, 466 f.
Cultelli, 443
Cuneus, 539, 540
Cupesidae, 234
Curculionidae, 290
Curtice, on Hypoderma, 516
Curupira, 465
Cut-worms, 415
Cyathoceridae, 243
Cybister laterimarginalis or roeseli, 210;
C. tripunctatus, 211
Cybocephalus, 232
Cyclica, 279, 282
Cyclorrhapha, 454;
C. Aschiza, 455, 494 f.;
C. Schizophora, 456, 503 f.
Cylidrus, 253
Cylindrotomina, 474
Cymatophoridae, 368, 386
Cymbidae, 410
Cynomyia mortuorum, 510
Cyphagogus segnipes, 296
Cyphanta, 368 n.
Cyphonia clavata, 576
Cyphonidae, 255
Cyrtidae, 489
Cyrtocorides, 545
Cyrtocoris monstrosus, 546

Dacnides, 237
Dactylopius citri, 595
Daddy-long-legs, 471
Dakruma coccidivora, 424
Dalla Torre, Catalogue of Hymenoptera, 21
Danaides, 344, 347
Danaioid Heliconiidae, 346
Danais archippus, or plexippus, 345
Dances, 351, 464, 493, 554
Darwin, C., on Pelobius, 208
Darwin, F., on proboscis of Lepidoptera, 311 n.
Dascillidae, 243, 255
Dascillus cervinus, 255
Dasychira pudibunda, 408;
D. rossii, 407
Dasygastres, 20, 35 f.
Dasypoda hirtipes, 27