Humanoid robot detection using deep learning

A speed-accuracy tradeoff

Mohammad Javadi, Sina Mokhtarzadeh Azar, Sajjad Azami, Saeed Shiry Ghidary, Soroush Sadeghnejad, Jacky Baltes

Research output: Chapter in Book/Report/Conference proceedingConference contribution

5 Citations (Scopus)

Abstract

Recent advances in computer vision have made the detection of landmarks on the soccer field easier for teams. However, the detection of other robots is also a critical capability that has not garnered much attention in the RoboCup community so far. This problem is well represented in different RoboCup Soccer and Rescue Robot Leagues. In this paper, we compare several two-stage detection systems based on various Convolutional Neural Networks (CNN) and highlight their speed-accuracy trade off. The approach performs edge based image segmentation in order to reduce the search space and then a CNN validates the detection in the second stage. We use images of different humanoid robots to train and test three different CNN architectures. A part of these images was gathered by our team and will be publicly available. Our experiments demonstrate the strong adaptability of deeper CNNs. These models, trained on a limited set of robots, are able to successfully distinguish an unseen kind of humanoid robot from non-robot regions.

Original languageEnglish
Title of host publicationRoboCup 2017
Subtitle of host publicationRobot World Cup XXI
EditorsHidehisa Akiyama, Oliver Obst, Claude Sammut, Flavio Tonidandel
PublisherSpringer Verlag
Pages338-349
Number of pages12
ISBN (Print)9783030003074
DOIs
Publication statusPublished - 2018 Jan 1
Event21st RoboCup International Symposium, 2017 - Nagoya, Japan
Duration: 2017 Jul 272017 Jul 31

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11175 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference21st RoboCup International Symposium, 2017
CountryJapan
CityNagoya
Period17/7/2717/7/31

Fingerprint

Humanoid Robot
Trade-offs
Robots
Robot
Neural Networks
Neural networks
Network Architecture
Landmarks
Adaptability
Image Segmentation
Computer Vision
Search Space
Network architecture
Image segmentation
Computer vision
Learning
Deep learning
Demonstrate
Experiment
Experiments

Keywords

  • Convolutional neural networks
  • Deep learning
  • Humanoid robots
  • Image segmentation
  • Robot detection
  • Robot vision

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

Javadi, M., Azar, S. M., Azami, S., Ghidary, S. S., Sadeghnejad, S., & Baltes, J. (2018). Humanoid robot detection using deep learning: A speed-accuracy tradeoff. In H. Akiyama, O. Obst, C. Sammut, & F. Tonidandel (Eds.), RoboCup 2017: Robot World Cup XXI (pp. 338-349). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11175 LNAI). Springer Verlag. https://doi.org/10.1007/978-3-030-00308-1_28

Humanoid robot detection using deep learning : A speed-accuracy tradeoff. / Javadi, Mohammad; Azar, Sina Mokhtarzadeh; Azami, Sajjad; Ghidary, Saeed Shiry; Sadeghnejad, Soroush; Baltes, Jacky.

RoboCup 2017: Robot World Cup XXI. ed. / Hidehisa Akiyama; Oliver Obst; Claude Sammut; Flavio Tonidandel. Springer Verlag, 2018. p. 338-349 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11175 LNAI).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Javadi, M, Azar, SM, Azami, S, Ghidary, SS, Sadeghnejad, S & Baltes, J 2018, Humanoid robot detection using deep learning: A speed-accuracy tradeoff. in H Akiyama, O Obst, C Sammut & F Tonidandel (eds), RoboCup 2017: Robot World Cup XXI. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11175 LNAI, Springer Verlag, pp. 338-349, 21st RoboCup International Symposium, 2017, Nagoya, Japan, 17/7/27. https://doi.org/10.1007/978-3-030-00308-1_28
Javadi M, Azar SM, Azami S, Ghidary SS, Sadeghnejad S, Baltes J. Humanoid robot detection using deep learning: A speed-accuracy tradeoff. In Akiyama H, Obst O, Sammut C, Tonidandel F, editors, RoboCup 2017: Robot World Cup XXI. Springer Verlag. 2018. p. 338-349. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-030-00308-1_28
Javadi, Mohammad ; Azar, Sina Mokhtarzadeh ; Azami, Sajjad ; Ghidary, Saeed Shiry ; Sadeghnejad, Soroush ; Baltes, Jacky. / Humanoid robot detection using deep learning : A speed-accuracy tradeoff. RoboCup 2017: Robot World Cup XXI. editor / Hidehisa Akiyama ; Oliver Obst ; Claude Sammut ; Flavio Tonidandel. Springer Verlag, 2018. pp. 338-349 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{0101b99f7a474e59afc0212156673ebf,
title = "Humanoid robot detection using deep learning: A speed-accuracy tradeoff",
abstract = "Recent advances in computer vision have made the detection of landmarks on the soccer field easier for teams. However, the detection of other robots is also a critical capability that has not garnered much attention in the RoboCup community so far. This problem is well represented in different RoboCup Soccer and Rescue Robot Leagues. In this paper, we compare several two-stage detection systems based on various Convolutional Neural Networks (CNN) and highlight their speed-accuracy trade off. The approach performs edge based image segmentation in order to reduce the search space and then a CNN validates the detection in the second stage. We use images of different humanoid robots to train and test three different CNN architectures. A part of these images was gathered by our team and will be publicly available. Our experiments demonstrate the strong adaptability of deeper CNNs. These models, trained on a limited set of robots, are able to successfully distinguish an unseen kind of humanoid robot from non-robot regions.",
keywords = "Convolutional neural networks, Deep learning, Humanoid robots, Image segmentation, Robot detection, Robot vision",
author = "Mohammad Javadi and Azar, {Sina Mokhtarzadeh} and Sajjad Azami and Ghidary, {Saeed Shiry} and Soroush Sadeghnejad and Jacky Baltes",
year = "2018",
month = "1",
day = "1",
doi = "10.1007/978-3-030-00308-1_28",
language = "English",
isbn = "9783030003074",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "338--349",
editor = "Hidehisa Akiyama and Oliver Obst and Claude Sammut and Flavio Tonidandel",
booktitle = "RoboCup 2017",

}

TY - GEN

T1 - Humanoid robot detection using deep learning

T2 - A speed-accuracy tradeoff

AU - Javadi, Mohammad

AU - Azar, Sina Mokhtarzadeh

AU - Azami, Sajjad

AU - Ghidary, Saeed Shiry

AU - Sadeghnejad, Soroush

AU - Baltes, Jacky

PY - 2018/1/1

Y1 - 2018/1/1

N2 - Recent advances in computer vision have made the detection of landmarks on the soccer field easier for teams. However, the detection of other robots is also a critical capability that has not garnered much attention in the RoboCup community so far. This problem is well represented in different RoboCup Soccer and Rescue Robot Leagues. In this paper, we compare several two-stage detection systems based on various Convolutional Neural Networks (CNN) and highlight their speed-accuracy trade off. The approach performs edge based image segmentation in order to reduce the search space and then a CNN validates the detection in the second stage. We use images of different humanoid robots to train and test three different CNN architectures. A part of these images was gathered by our team and will be publicly available. Our experiments demonstrate the strong adaptability of deeper CNNs. These models, trained on a limited set of robots, are able to successfully distinguish an unseen kind of humanoid robot from non-robot regions.

AB - Recent advances in computer vision have made the detection of landmarks on the soccer field easier for teams. However, the detection of other robots is also a critical capability that has not garnered much attention in the RoboCup community so far. This problem is well represented in different RoboCup Soccer and Rescue Robot Leagues. In this paper, we compare several two-stage detection systems based on various Convolutional Neural Networks (CNN) and highlight their speed-accuracy trade off. The approach performs edge based image segmentation in order to reduce the search space and then a CNN validates the detection in the second stage. We use images of different humanoid robots to train and test three different CNN architectures. A part of these images was gathered by our team and will be publicly available. Our experiments demonstrate the strong adaptability of deeper CNNs. These models, trained on a limited set of robots, are able to successfully distinguish an unseen kind of humanoid robot from non-robot regions.

KW - Convolutional neural networks

KW - Deep learning

KW - Humanoid robots

KW - Image segmentation

KW - Robot detection

KW - Robot vision

UR - http://www.scopus.com/inward/record.url?scp=85053941550&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85053941550&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-00308-1_28

DO - 10.1007/978-3-030-00308-1_28

M3 - Conference contribution

SN - 9783030003074

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 338

EP - 349

BT - RoboCup 2017

A2 - Akiyama, Hidehisa

A2 - Obst, Oliver

A2 - Sammut, Claude

A2 - Tonidandel, Flavio

PB - Springer Verlag

ER -