Enhancing object detection for autonomous driving by optimizing anchor generation and addressing class imbalance

Object detection has been one of the most active topics in computer vision for the past years. Recent works have mainly focused on pushing the state-of-the-art in the general-purpose COCO benchmark. However, the use of such detection frameworks in specific applications such as autonomous driving is...

Descripción completa

Detalles Bibliográficos
Autores: Carranza García, Manuel, Lara Benítez, Pedro, García Gutiérrez, Jorge, Riquelme Santos, José Cristóbal
Tipo de recurso: artículo
Estado:Versión publicada
Fecha de publicación:2021
País:España
Institución:Universidad de Sevilla (US)
Repositorio:idUS. Depósito de Investigación de la Universidad de Sevilla
OAI Identifier:oai:idus.us.es:11441/130039
Acceso en línea:https://hdl.handle.net/11441/130039
https://doi.org/10.1016/j.neucom.2021.04.001
Access Level:acceso abierto
Palabra clave:Autonomous vehicles
Anchor optimization
Class imbalance
Convolutional Neural Network
Deep learning
Object detection
id ES_b14c4e871e2988a2d7255db65daea0c4
oai_identifier_str oai:idus.us.es:11441/130039
network_acronym_str ES
network_name_str España
repository_id_str
spelling Enhancing object detection for autonomous driving by optimizing anchor generation and addressing class imbalanceCarranza García, ManuelLara Benítez, PedroGarcía Gutiérrez, JorgeRiquelme Santos, José CristóbalAutonomous vehiclesAnchor optimizationClass imbalanceConvolutional Neural NetworkDeep learningObject detectionObject detection has been one of the most active topics in computer vision for the past years. Recent works have mainly focused on pushing the state-of-the-art in the general-purpose COCO benchmark. However, the use of such detection frameworks in specific applications such as autonomous driving is yet an area to be addressed. This study presents an enhanced 2D object detector based on Faster RCNN that is better suited for the context of autonomous vehicles. Two main aspects are improved: the anchor generation procedure and the performance drop in minority classes. The default uniform anchor configuration is not suitable in this scenario due to the perspective projection of the vehicle cameras. Therefore, we propose a perspective-aware methodology that divides the image into key regions via clustering and uses evolutionary algorithms to optimize the base anchors for each of them. Furthermore, we add a module that enhances the precision of the second-stage header network by including the spatial information of the candidate regions proposed in the first stage. We also explore different reweighting strategies to address the foreground-foreground class imbalance, showing that the use of a reduced version of focal loss can significantly improve the detection of difficult and underrepresented objects in two-stage detectors. Finally, we design an ensemble model to combine the strengths of the different learning strategies. Our proposal is evaluated with the Waymo Open Dataset, which is the most extensive and diverse up to date. The results demonstrate an average accuracy improvement of 6.13% mAP when using the best single model, and of 9.69% mAP with the ensemble. The proposed modifications over the Faster R-CNN do not increase computational cost and can easily be extended to optimize other anchor-based detection frameworks.Ministerio de Ciencia, Innovación y Universidades TIN2017-88209-C2Junta de Andalucía US-1263341Junta de Andalucía P18-RT-2778ElsevierLenguajes y Sistemas InformáticosMinisterio de Ciencia, Innovación y Universidades (MICINN). EspañaJunta de Andalucía2021info:eu-repo/semantics/articleinfo:eu-repo/semantics/publishedVersionapplication/pdfapplication/pdfhttps://hdl.handle.net/11441/130039https://doi.org/10.1016/j.neucom.2021.04.001reponame:idUS. Depósito de Investigación de la Universidad de Sevillainstname:Universidad de Sevilla (US)InglésNeurocomputing, 449 (August 2021), 229-244.TIN2017-88209-C2US-1263341P18-RT-2778https://www.sciencedirect.com/science/article/pii/S0925231221005191?via%3Dihubinfo:eu-repo/semantics/openAccessoai:idus.us.es:11441/1300392026-06-17T12:51:07Z
dc.title.none.fl_str_mv Enhancing object detection for autonomous driving by optimizing anchor generation and addressing class imbalance
title Enhancing object detection for autonomous driving by optimizing anchor generation and addressing class imbalance
spellingShingle Enhancing object detection for autonomous driving by optimizing anchor generation and addressing class imbalance
Carranza García, Manuel
Autonomous vehicles
Anchor optimization
Class imbalance
Convolutional Neural Network
Deep learning
Object detection
title_short Enhancing object detection for autonomous driving by optimizing anchor generation and addressing class imbalance
title_full Enhancing object detection for autonomous driving by optimizing anchor generation and addressing class imbalance
title_fullStr Enhancing object detection for autonomous driving by optimizing anchor generation and addressing class imbalance
title_full_unstemmed Enhancing object detection for autonomous driving by optimizing anchor generation and addressing class imbalance
title_sort Enhancing object detection for autonomous driving by optimizing anchor generation and addressing class imbalance
dc.creator.none.fl_str_mv Carranza García, Manuel
Lara Benítez, Pedro
García Gutiérrez, Jorge
Riquelme Santos, José Cristóbal
author Carranza García, Manuel
author_facet Carranza García, Manuel
Lara Benítez, Pedro
García Gutiérrez, Jorge
Riquelme Santos, José Cristóbal
author_role author
author2 Lara Benítez, Pedro
García Gutiérrez, Jorge
Riquelme Santos, José Cristóbal
author2_role author
author
author
dc.contributor.none.fl_str_mv Lenguajes y Sistemas Informáticos
Ministerio de Ciencia, Innovación y Universidades (MICINN). España
Junta de Andalucía
dc.subject.none.fl_str_mv Autonomous vehicles
Anchor optimization
Class imbalance
Convolutional Neural Network
Deep learning
Object detection
topic Autonomous vehicles
Anchor optimization
Class imbalance
Convolutional Neural Network
Deep learning
Object detection
description Object detection has been one of the most active topics in computer vision for the past years. Recent works have mainly focused on pushing the state-of-the-art in the general-purpose COCO benchmark. However, the use of such detection frameworks in specific applications such as autonomous driving is yet an area to be addressed. This study presents an enhanced 2D object detector based on Faster RCNN that is better suited for the context of autonomous vehicles. Two main aspects are improved: the anchor generation procedure and the performance drop in minority classes. The default uniform anchor configuration is not suitable in this scenario due to the perspective projection of the vehicle cameras. Therefore, we propose a perspective-aware methodology that divides the image into key regions via clustering and uses evolutionary algorithms to optimize the base anchors for each of them. Furthermore, we add a module that enhances the precision of the second-stage header network by including the spatial information of the candidate regions proposed in the first stage. We also explore different reweighting strategies to address the foreground-foreground class imbalance, showing that the use of a reduced version of focal loss can significantly improve the detection of difficult and underrepresented objects in two-stage detectors. Finally, we design an ensemble model to combine the strengths of the different learning strategies. Our proposal is evaluated with the Waymo Open Dataset, which is the most extensive and diverse up to date. The results demonstrate an average accuracy improvement of 6.13% mAP when using the best single model, and of 9.69% mAP with the ensemble. The proposed modifications over the Faster R-CNN do not increase computational cost and can easily be extended to optimize other anchor-based detection frameworks.
publishDate 2021
dc.date.none.fl_str_mv 2021
dc.type.none.fl_str_mv info:eu-repo/semantics/article
info:eu-repo/semantics/publishedVersion
format article
status_str publishedVersion
dc.identifier.none.fl_str_mv https://hdl.handle.net/11441/130039
https://doi.org/10.1016/j.neucom.2021.04.001
url https://hdl.handle.net/11441/130039
https://doi.org/10.1016/j.neucom.2021.04.001
dc.language.none.fl_str_mv Inglés
language_invalid_str_mv Inglés
dc.relation.none.fl_str_mv Neurocomputing, 449 (August 2021), 229-244.
TIN2017-88209-C2
US-1263341
P18-RT-2778
https://www.sciencedirect.com/science/article/pii/S0925231221005191?via%3Dihub
dc.rights.none.fl_str_mv info:eu-repo/semantics/openAccess
eu_rights_str_mv openAccess
dc.format.none.fl_str_mv application/pdf
application/pdf
dc.publisher.none.fl_str_mv Elsevier
publisher.none.fl_str_mv Elsevier
dc.source.none.fl_str_mv reponame:idUS. Depósito de Investigación de la Universidad de Sevilla
instname:Universidad de Sevilla (US)
instname_str Universidad de Sevilla (US)
reponame_str idUS. Depósito de Investigación de la Universidad de Sevilla
collection idUS. Depósito de Investigación de la Universidad de Sevilla
repository.name.fl_str_mv
repository.mail.fl_str_mv
_version_ 1869416918751903744
score 15,300724

Similares en LA Referencia