The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
La détection des piétons est une tâche essentielle mais difficile en vision par ordinateur, en particulier dans les scènes bondées en raison d'une forte occlusion intra-classe. Dans le système visuel humain, les informations sur la tête peuvent être utilisées pour localiser un piéton dans une foule, car elles sont plus stables et moins susceptibles d'être obstruées. Inspirés par cet indice, nous proposons un détecteur double tâche qui détecte simultanément la tête et le corps humain. Concrètement, nous estimons les candidats corps humains à partir des régions de la tête avec un rapport tête-corps statistique. UN tête-corps une carte d'alignement est proposée pour effectuer un apprentissage relationnel entre les corps humains et les têtes en fonction de leur corrélation inhérente. Nous exploitons les informations de tête comme critère de détection strict pour supprimer les faux positifs courants de la détection des piétons via une nouvelle perte pull-push. Nous validons l'efficacité de la méthode proposée sur les benchmarks CrowdHuman et CityPersons. Les résultats expérimentaux démontrent que la méthode proposée atteint des performances impressionnantes dans la détection des piétons fortement obstrués avec peu de coûts de calcul supplémentaires.
Chen CHEN
National University of Defense Technology
Maojun ZHANG
National University of Defense Technology
Hanlin TAN
National University of Defense Technology
Huaxin XIAO
National University of Defense Technology
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copier
Chen CHEN, Maojun ZHANG, Hanlin TAN, Huaxin XIAO, "Co-Head Pedestrian Detection in Crowded Scenes" in IEICE TRANSACTIONS on Fundamentals,
vol. E104-A, no. 10, pp. 1440-1444, October 2021, doi: 10.1587/transfun.2020EAL2110.
Abstract: Pedestrian detection is an essential but challenging task in computer vision, especially in crowded scenes due to heavy intra-class occlusion. In human visual system, head information can be used to locate pedestrian in a crowd because it is more stable and less likely to be occluded. Inspired by this clue, we propose a dual-task detector which detects head and human body simultaneously. Concretely, we estimate human body candidates from head regions with statistical head-body ratio. A head-body alignment map is proposed to perform relational learning between human bodies and heads based on their inherent correlation. We leverage the head information as a strict detection criterion to suppress common false positives of pedestrian detection via a novel pull-push loss. We validate the effectiveness of the proposed method on the CrowdHuman and CityPersons benchmarks. Experimental results demonstrate that the proposed method achieves impressive performance in detecting heavy-occluded pedestrians with little additional computation cost.
URL: https://global.ieice.org/en_transactions/fundamentals/10.1587/transfun.2020EAL2110/_p
Copier
@ARTICLE{e104-a_10_1440,
author={Chen CHEN, Maojun ZHANG, Hanlin TAN, Huaxin XIAO, },
journal={IEICE TRANSACTIONS on Fundamentals},
title={Co-Head Pedestrian Detection in Crowded Scenes},
year={2021},
volume={E104-A},
number={10},
pages={1440-1444},
abstract={Pedestrian detection is an essential but challenging task in computer vision, especially in crowded scenes due to heavy intra-class occlusion. In human visual system, head information can be used to locate pedestrian in a crowd because it is more stable and less likely to be occluded. Inspired by this clue, we propose a dual-task detector which detects head and human body simultaneously. Concretely, we estimate human body candidates from head regions with statistical head-body ratio. A head-body alignment map is proposed to perform relational learning between human bodies and heads based on their inherent correlation. We leverage the head information as a strict detection criterion to suppress common false positives of pedestrian detection via a novel pull-push loss. We validate the effectiveness of the proposed method on the CrowdHuman and CityPersons benchmarks. Experimental results demonstrate that the proposed method achieves impressive performance in detecting heavy-occluded pedestrians with little additional computation cost.},
keywords={},
doi={10.1587/transfun.2020EAL2110},
ISSN={1745-1337},
month={October},}
Copier
TY - JOUR
TI - Co-Head Pedestrian Detection in Crowded Scenes
T2 - IEICE TRANSACTIONS on Fundamentals
SP - 1440
EP - 1444
AU - Chen CHEN
AU - Maojun ZHANG
AU - Hanlin TAN
AU - Huaxin XIAO
PY - 2021
DO - 10.1587/transfun.2020EAL2110
JO - IEICE TRANSACTIONS on Fundamentals
SN - 1745-1337
VL - E104-A
IS - 10
JA - IEICE TRANSACTIONS on Fundamentals
Y1 - October 2021
AB - Pedestrian detection is an essential but challenging task in computer vision, especially in crowded scenes due to heavy intra-class occlusion. In human visual system, head information can be used to locate pedestrian in a crowd because it is more stable and less likely to be occluded. Inspired by this clue, we propose a dual-task detector which detects head and human body simultaneously. Concretely, we estimate human body candidates from head regions with statistical head-body ratio. A head-body alignment map is proposed to perform relational learning between human bodies and heads based on their inherent correlation. We leverage the head information as a strict detection criterion to suppress common false positives of pedestrian detection via a novel pull-push loss. We validate the effectiveness of the proposed method on the CrowdHuman and CityPersons benchmarks. Experimental results demonstrate that the proposed method achieves impressive performance in detecting heavy-occluded pedestrians with little additional computation cost.
ER -