Presentation 2010-03-05
A Study on Improving Human Detection Using HOG Descriptor
Yusuke MURASE, Sho MIURA, Hiroyuki TSUJI, Tomoaki KIMURA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) We studied Dalal's algorithm, one of the most reliable human detection algorithm that uses HOG features to classify detection windows into human or non-human category with SVM, and made some proposals to improve its detection ability. We first adjusted the ratio of positive and negative window and then introduced window map operation to reduce a large number of misdetected windows as human. Experimental results show that the proposed method is effective to eliminate misdetected windows and to choose a single window from multiply detected windows for the same human object.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Human Detection / HOG Descriptor / SVM / Window Map / Connected Component Labeling
Paper # SIS2009-68
Date of Issue

Conference Information
Committee SIS
Conference Date 2010/2/25(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Smart Info-Media Systems (SIS)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) A Study on Improving Human Detection Using HOG Descriptor
Sub Title (in English)
Keyword(1) Human Detection
Keyword(2) HOG Descriptor
Keyword(3) SVM
Keyword(4) Window Map
Keyword(5) Connected Component Labeling
1st Author's Name Yusuke MURASE
1st Author's Affiliation Faculty of Information Technology, Kanagawa Institute of Technology()
2nd Author's Name Sho MIURA
2nd Author's Affiliation Faculty of Information Technology, Kanagawa Institute of Technology
3rd Author's Name Hiroyuki TSUJI
3rd Author's Affiliation Faculty of Information Technology, Kanagawa Institute of Technology
4th Author's Name Tomoaki KIMURA
4th Author's Affiliation Faculty of Information Technology, Kanagawa Institute of Technology
Date 2010-03-05
Paper # SIS2009-68
Volume (vol) vol.109
Number (no) 447
Page pp.pp.-
#Pages 6
Date of Issue