Humanoid robot detection using deep learning: A speed-accuracy tradeoff

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

*Corresponding author for this work

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

15 Citations (Scopus)


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
Number of pages12
ISBN (Print)9783030003074
Publication statusPublished - 2018
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


Conference21st RoboCup International Symposium, 2017


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

ASJC Scopus subject areas

  • Theoretical Computer Science
  • General Computer Science


Dive into the research topics of 'Humanoid robot detection using deep learning: A speed-accuracy tradeoff'. Together they form a unique fingerprint.

Cite this