Human detection by a small autonomous mobile robot

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

Abstract

We propose a human detection method using HOG and SVM from an image by a small autonomous mobile robot. Existing works for human detection from images cannot be used for our purpose because they assume that the images are taken from a high position, at least at the height of a small human child, while our robot is of 15cm height. The combination of HOG and SVM is known as the most successful human detection method so we adopt it. To cope with a wide variety of human shapes mainly due to the distance to them, we devised a two stage prediction method which uses two kinds of SVM classifiers based on an estimation of the distance. The estimation is based on the ratio of the skin color pixel in the image, which allows us to clearly separate our problem into whole body detection and partial body detection. Experiments in an office showed promising results of our method with F value 0.93.

Original languageEnglish
Title of host publicationExtraction et Gestion des Connaissances, EGC 2012
Pages233-238
Number of pages6
Publication statusPublished - Dec 1 2012
Event12th Journees Internationales Francophones Extraction et Gestion des Connaissances, EGC 2012 - 12th French-speaking International Conference on Knowledge Discovery and Management, EGC 2012 - Bordeaux, France
Duration: Jan 31 2012Feb 3 2012

Publication series

NameRevue des Nouvelles Technologies de l'Information
VolumeE.23
ISSN (Print)1764-1667

Other

Other12th Journees Internationales Francophones Extraction et Gestion des Connaissances, EGC 2012 - 12th French-speaking International Conference on Knowledge Discovery and Management, EGC 2012
CountryFrance
CityBordeaux
Period1/31/122/3/12

Fingerprint

Mobile robots
Skin
Classifiers
Pixels
Robots
Color
Experiments

All Science Journal Classification (ASJC) codes

  • Computer Networks and Communications
  • Computer Science Applications
  • Information Systems
  • Software

Cite this

Takemoto, K., Takano, S., & Suzuki, E. (2012). Human detection by a small autonomous mobile robot. In Extraction et Gestion des Connaissances, EGC 2012 (pp. 233-238). (Revue des Nouvelles Technologies de l'Information; Vol. E.23).

Human detection by a small autonomous mobile robot. / Takemoto, Kouhei; Takano, Shigeru; Suzuki, Einoshin.

Extraction et Gestion des Connaissances, EGC 2012. 2012. p. 233-238 (Revue des Nouvelles Technologies de l'Information; Vol. E.23).

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

Takemoto, K, Takano, S & Suzuki, E 2012, Human detection by a small autonomous mobile robot. in Extraction et Gestion des Connaissances, EGC 2012. Revue des Nouvelles Technologies de l'Information, vol. E.23, pp. 233-238, 12th Journees Internationales Francophones Extraction et Gestion des Connaissances, EGC 2012 - 12th French-speaking International Conference on Knowledge Discovery and Management, EGC 2012, Bordeaux, France, 1/31/12.
Takemoto K, Takano S, Suzuki E. Human detection by a small autonomous mobile robot. In Extraction et Gestion des Connaissances, EGC 2012. 2012. p. 233-238. (Revue des Nouvelles Technologies de l'Information).
Takemoto, Kouhei ; Takano, Shigeru ; Suzuki, Einoshin. / Human detection by a small autonomous mobile robot. Extraction et Gestion des Connaissances, EGC 2012. 2012. pp. 233-238 (Revue des Nouvelles Technologies de l'Information).
@inproceedings{592ed070117549739837efd357579bdf,
title = "Human detection by a small autonomous mobile robot",
abstract = "We propose a human detection method using HOG and SVM from an image by a small autonomous mobile robot. Existing works for human detection from images cannot be used for our purpose because they assume that the images are taken from a high position, at least at the height of a small human child, while our robot is of 15cm height. The combination of HOG and SVM is known as the most successful human detection method so we adopt it. To cope with a wide variety of human shapes mainly due to the distance to them, we devised a two stage prediction method which uses two kinds of SVM classifiers based on an estimation of the distance. The estimation is based on the ratio of the skin color pixel in the image, which allows us to clearly separate our problem into whole body detection and partial body detection. Experiments in an office showed promising results of our method with F value 0.93.",
author = "Kouhei Takemoto and Shigeru Takano and Einoshin Suzuki",
year = "2012",
month = "12",
day = "1",
language = "English",
isbn = "9782705683108",
series = "Revue des Nouvelles Technologies de l'Information",
pages = "233--238",
booktitle = "Extraction et Gestion des Connaissances, EGC 2012",

}

TY - GEN

T1 - Human detection by a small autonomous mobile robot

AU - Takemoto, Kouhei

AU - Takano, Shigeru

AU - Suzuki, Einoshin

PY - 2012/12/1

Y1 - 2012/12/1

N2 - We propose a human detection method using HOG and SVM from an image by a small autonomous mobile robot. Existing works for human detection from images cannot be used for our purpose because they assume that the images are taken from a high position, at least at the height of a small human child, while our robot is of 15cm height. The combination of HOG and SVM is known as the most successful human detection method so we adopt it. To cope with a wide variety of human shapes mainly due to the distance to them, we devised a two stage prediction method which uses two kinds of SVM classifiers based on an estimation of the distance. The estimation is based on the ratio of the skin color pixel in the image, which allows us to clearly separate our problem into whole body detection and partial body detection. Experiments in an office showed promising results of our method with F value 0.93.

AB - We propose a human detection method using HOG and SVM from an image by a small autonomous mobile robot. Existing works for human detection from images cannot be used for our purpose because they assume that the images are taken from a high position, at least at the height of a small human child, while our robot is of 15cm height. The combination of HOG and SVM is known as the most successful human detection method so we adopt it. To cope with a wide variety of human shapes mainly due to the distance to them, we devised a two stage prediction method which uses two kinds of SVM classifiers based on an estimation of the distance. The estimation is based on the ratio of the skin color pixel in the image, which allows us to clearly separate our problem into whole body detection and partial body detection. Experiments in an office showed promising results of our method with F value 0.93.

UR - http://www.scopus.com/inward/record.url?scp=84877273790&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84877273790&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:84877273790

SN - 9782705683108

T3 - Revue des Nouvelles Technologies de l'Information

SP - 233

EP - 238

BT - Extraction et Gestion des Connaissances, EGC 2012

ER -