Rastreamento em tempo real de múltiplos objetos por associação de detecções
Autor(a) principal: | |
---|---|
Data de Publicação: | 2019 |
Tipo de documento: | Dissertação |
Idioma: | por |
Título da fonte: | Repositório Institucional da UFS |
Texto Completo: | https://ri.ufs.br/jspui/handle/riufs/14548 |
Resumo: | With the recent advances in the object detection research field, tracking-by-detection has become the leading paradigm adopted by multi-object tracking algorithms. By extracting different features from detected objects, those algorithms can estimate the objects’ similarities and association patterns along successive frames. However, since similarity functions applied by tracking algorithms are handcrafted, it is difficult to employ them in new contexts. In this study, it is investigated the use of artificial neural networks to automatically learning a similarity function that can be used among detections. During training, the networks were introduced to correct and incorrect association patterns, sampled from real tracking datasets. For such, different motion and appearance features have been considered. A trained network has been inserted into a multiple-object tracking framework, which has been assessed on three different experiment scenarios: the tracking of pedestrians, the tracking of bus passengers and the automatic counting of bus passengers. During the first, conducted on the MOT Challenge benchmark, the proposed method scored an accuracy similar to state-of-the-art trackers, but at a computational cost at least 16 times lower. In the second experiment, which was based on a local dataset, the proposed tracker matched the results obtained by its direct baseline, the DeepSORT tracker, but with a speed gain of 42.8%. Finally, the third experiment was a study case, where the median absolute error scored by the proposed method was 40.7% inferior to its baseline. In the end, this study could demonstrate that the proposal method can be automatically adapted to different tracking scenarios while presenting highly competitive cost-effectiveness when compared to those algorithms considered in the experiments. |
id |
UFS-2_5e4974616cd7c7119047ff1ee9d1a850 |
---|---|
oai_identifier_str |
oai:ufs.br:riufs/14548 |
network_acronym_str |
UFS-2 |
network_name_str |
Repositório Institucional da UFS |
repository_id_str |
|
spelling |
Meneses, Michel Conrado CardosoMatos, Leonardo NogueiraPrado, Bruno Otavio Piedade2021-09-01T18:27:50Z2021-09-01T18:27:50Z2019-10-25MENESES, Michel Conrado Cardoso. Rastreamento em tempo real de múltiplos objetos por associação de detecções. 2019. 136 f. Dissertação (Mestrado em Ciência da Computação) - Universidade Federal de Sergipe, São Cristóvão, Sergipe, 2019.https://ri.ufs.br/jspui/handle/riufs/14548With the recent advances in the object detection research field, tracking-by-detection has become the leading paradigm adopted by multi-object tracking algorithms. By extracting different features from detected objects, those algorithms can estimate the objects’ similarities and association patterns along successive frames. However, since similarity functions applied by tracking algorithms are handcrafted, it is difficult to employ them in new contexts. In this study, it is investigated the use of artificial neural networks to automatically learning a similarity function that can be used among detections. During training, the networks were introduced to correct and incorrect association patterns, sampled from real tracking datasets. For such, different motion and appearance features have been considered. A trained network has been inserted into a multiple-object tracking framework, which has been assessed on three different experiment scenarios: the tracking of pedestrians, the tracking of bus passengers and the automatic counting of bus passengers. During the first, conducted on the MOT Challenge benchmark, the proposed method scored an accuracy similar to state-of-the-art trackers, but at a computational cost at least 16 times lower. In the second experiment, which was based on a local dataset, the proposed tracker matched the results obtained by its direct baseline, the DeepSORT tracker, but with a speed gain of 42.8%. Finally, the third experiment was a study case, where the median absolute error scored by the proposed method was 40.7% inferior to its baseline. In the end, this study could demonstrate that the proposal method can be automatically adapted to different tracking scenarios while presenting highly competitive cost-effectiveness when compared to those algorithms considered in the experiments.Devido ao recente avanço na área de detecção de objetos, o rastreamento por detecção (no inglês, tracking-by-detection) tornou-se o principal paradigma adotado por rastreadores de múltiplos objetos. Com base na extração de diferentes características dos objetos detectados, tais algoritmos são capazes de estimar a similaridade e o padrão de associação dos objetos ao longo de sucessivas imagens. No entanto, uma vez que as funções de similaridade aplicadas por algoritmos de rastreamento são construídas manualmente, sua adaptação para novos cenários é dificultada. Este trabalho investigou o uso de técnicas de aprendizado de máquina baseadas em redes neurais artificiais para a indução automática de funções de similaridade entre objetos. Durante seu treinamento, tais redes foram apresentadas a padrões de associações corretas e incorretas entre detecções de objetos amostradas de bases reais de rastreamento. Para tanto, diferentes características relacionadas à aparência e à movimentação de objetos foram consideradas. Uma rede treinada foi inserida num framework de rastreamento de múltiplos objetos, o qual foi avaliado em três diferentes cenários de experimentação: o rastreamento de pedestres, o rastreamento de passageiros de ônibus e a contagem automática de passageiros de ônibus. No primeiro experimento, realizado com base no benchmark MOT Challenge, o método proposto obteve acurácia similar à apresentada por algoritmos considerados estado da arte, porém a um custo computacional até 16 vezes menor. Já o segundo experimento foi realizado a partir de uma base de dados construída localmente, sobre a qual o método proposto igualou a acurácia de sua principal baseline, o método DeepSORT, porém com ganho de 42, 8% em velocidade. O terceiro experimento correspondeu a um estudo de caso no qual a contagem obtida através do método proposto apresentou um erro médio absoluto 40, 7% menor que sua baseline. Ao fim deste trabalho foi possível verificar que o método proposto pode ser adaptado automaticamente a diferentes cenários e apresenta competitiva relação acurácia vs velocidade de execução, quando comparado aos algoritmos considerados nos experimentos.Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPESSão Cristóvão, SEporRastreamento de múltiplos objetosVisão computacionalAprendizado de máquinaMultiple-object trackingComputer visionMachine learningCIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAORastreamento em tempo real de múltiplos objetos por associação de detecçõesinfo:eu-repo/semantics/publishedVersioninfo:eu-repo/semantics/masterThesisPós-Graduação em Ciência da ComputaçãoUniversidade Federal de Sergipereponame:Repositório Institucional da UFSinstname:Universidade Federal de Sergipe (UFS)instacron:UFSinfo:eu-repo/semantics/openAccessORIGINALMICHEL_CONRADO_CARDOSO_MENESES.pdfMICHEL_CONRADO_CARDOSO_MENESES.pdfapplication/pdf42100860https://ri.ufs.br/jspui/bitstream/riufs/14548/2/MICHEL_CONRADO_CARDOSO_MENESES.pdf7617d9edaf8b4a29c6695f2fa5655bf5MD52LICENSElicense.txtlicense.txttext/plain; charset=utf-81475https://ri.ufs.br/jspui/bitstream/riufs/14548/1/license.txt098cbbf65c2c15e1fb2e49c5d306a44cMD51TEXTMICHEL_CONRADO_CARDOSO_MENESES.pdf.txtMICHEL_CONRADO_CARDOSO_MENESES.pdf.txtExtracted texttext/plain266797https://ri.ufs.br/jspui/bitstream/riufs/14548/3/MICHEL_CONRADO_CARDOSO_MENESES.pdf.txt6bd90a6951597ccc8415630c3a5dc6b1MD53THUMBNAILMICHEL_CONRADO_CARDOSO_MENESES.pdf.jpgMICHEL_CONRADO_CARDOSO_MENESES.pdf.jpgGenerated Thumbnailimage/jpeg1435https://ri.ufs.br/jspui/bitstream/riufs/14548/4/MICHEL_CONRADO_CARDOSO_MENESES.pdf.jpgb5b8ff18bbfe4c9b169a4914741bf074MD54riufs/145482021-09-01 15:27:54.131oai:ufs.br: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Repositório InstitucionalPUBhttps://ri.ufs.br/oai/requestrepositorio@academico.ufs.bropendoar:2021-09-01T18:27:54Repositório Institucional da UFS - Universidade Federal de Sergipe (UFS)false |
dc.title.pt_BR.fl_str_mv |
Rastreamento em tempo real de múltiplos objetos por associação de detecções |
title |
Rastreamento em tempo real de múltiplos objetos por associação de detecções |
spellingShingle |
Rastreamento em tempo real de múltiplos objetos por associação de detecções Meneses, Michel Conrado Cardoso Rastreamento de múltiplos objetos Visão computacional Aprendizado de máquina Multiple-object tracking Computer vision Machine learning CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO |
title_short |
Rastreamento em tempo real de múltiplos objetos por associação de detecções |
title_full |
Rastreamento em tempo real de múltiplos objetos por associação de detecções |
title_fullStr |
Rastreamento em tempo real de múltiplos objetos por associação de detecções |
title_full_unstemmed |
Rastreamento em tempo real de múltiplos objetos por associação de detecções |
title_sort |
Rastreamento em tempo real de múltiplos objetos por associação de detecções |
author |
Meneses, Michel Conrado Cardoso |
author_facet |
Meneses, Michel Conrado Cardoso |
author_role |
author |
dc.contributor.author.fl_str_mv |
Meneses, Michel Conrado Cardoso |
dc.contributor.advisor1.fl_str_mv |
Matos, Leonardo Nogueira |
dc.contributor.advisor-co1.fl_str_mv |
Prado, Bruno Otavio Piedade |
contributor_str_mv |
Matos, Leonardo Nogueira Prado, Bruno Otavio Piedade |
dc.subject.por.fl_str_mv |
Rastreamento de múltiplos objetos Visão computacional Aprendizado de máquina |
topic |
Rastreamento de múltiplos objetos Visão computacional Aprendizado de máquina Multiple-object tracking Computer vision Machine learning CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO |
dc.subject.eng.fl_str_mv |
Multiple-object tracking Computer vision Machine learning |
dc.subject.cnpq.fl_str_mv |
CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO |
description |
With the recent advances in the object detection research field, tracking-by-detection has become the leading paradigm adopted by multi-object tracking algorithms. By extracting different features from detected objects, those algorithms can estimate the objects’ similarities and association patterns along successive frames. However, since similarity functions applied by tracking algorithms are handcrafted, it is difficult to employ them in new contexts. In this study, it is investigated the use of artificial neural networks to automatically learning a similarity function that can be used among detections. During training, the networks were introduced to correct and incorrect association patterns, sampled from real tracking datasets. For such, different motion and appearance features have been considered. A trained network has been inserted into a multiple-object tracking framework, which has been assessed on three different experiment scenarios: the tracking of pedestrians, the tracking of bus passengers and the automatic counting of bus passengers. During the first, conducted on the MOT Challenge benchmark, the proposed method scored an accuracy similar to state-of-the-art trackers, but at a computational cost at least 16 times lower. In the second experiment, which was based on a local dataset, the proposed tracker matched the results obtained by its direct baseline, the DeepSORT tracker, but with a speed gain of 42.8%. Finally, the third experiment was a study case, where the median absolute error scored by the proposed method was 40.7% inferior to its baseline. In the end, this study could demonstrate that the proposal method can be automatically adapted to different tracking scenarios while presenting highly competitive cost-effectiveness when compared to those algorithms considered in the experiments. |
publishDate |
2019 |
dc.date.issued.fl_str_mv |
2019-10-25 |
dc.date.accessioned.fl_str_mv |
2021-09-01T18:27:50Z |
dc.date.available.fl_str_mv |
2021-09-01T18:27:50Z |
dc.type.status.fl_str_mv |
info:eu-repo/semantics/publishedVersion |
dc.type.driver.fl_str_mv |
info:eu-repo/semantics/masterThesis |
format |
masterThesis |
status_str |
publishedVersion |
dc.identifier.citation.fl_str_mv |
MENESES, Michel Conrado Cardoso. Rastreamento em tempo real de múltiplos objetos por associação de detecções. 2019. 136 f. Dissertação (Mestrado em Ciência da Computação) - Universidade Federal de Sergipe, São Cristóvão, Sergipe, 2019. |
dc.identifier.uri.fl_str_mv |
https://ri.ufs.br/jspui/handle/riufs/14548 |
identifier_str_mv |
MENESES, Michel Conrado Cardoso. Rastreamento em tempo real de múltiplos objetos por associação de detecções. 2019. 136 f. Dissertação (Mestrado em Ciência da Computação) - Universidade Federal de Sergipe, São Cristóvão, Sergipe, 2019. |
url |
https://ri.ufs.br/jspui/handle/riufs/14548 |
dc.language.iso.fl_str_mv |
por |
language |
por |
dc.rights.driver.fl_str_mv |
info:eu-repo/semantics/openAccess |
eu_rights_str_mv |
openAccess |
dc.publisher.program.fl_str_mv |
Pós-Graduação em Ciência da Computação |
dc.publisher.initials.fl_str_mv |
Universidade Federal de Sergipe |
dc.source.none.fl_str_mv |
reponame:Repositório Institucional da UFS instname:Universidade Federal de Sergipe (UFS) instacron:UFS |
instname_str |
Universidade Federal de Sergipe (UFS) |
instacron_str |
UFS |
institution |
UFS |
reponame_str |
Repositório Institucional da UFS |
collection |
Repositório Institucional da UFS |
bitstream.url.fl_str_mv |
https://ri.ufs.br/jspui/bitstream/riufs/14548/2/MICHEL_CONRADO_CARDOSO_MENESES.pdf https://ri.ufs.br/jspui/bitstream/riufs/14548/1/license.txt https://ri.ufs.br/jspui/bitstream/riufs/14548/3/MICHEL_CONRADO_CARDOSO_MENESES.pdf.txt https://ri.ufs.br/jspui/bitstream/riufs/14548/4/MICHEL_CONRADO_CARDOSO_MENESES.pdf.jpg |
bitstream.checksum.fl_str_mv |
7617d9edaf8b4a29c6695f2fa5655bf5 098cbbf65c2c15e1fb2e49c5d306a44c 6bd90a6951597ccc8415630c3a5dc6b1 b5b8ff18bbfe4c9b169a4914741bf074 |
bitstream.checksumAlgorithm.fl_str_mv |
MD5 MD5 MD5 MD5 |
repository.name.fl_str_mv |
Repositório Institucional da UFS - Universidade Federal de Sergipe (UFS) |
repository.mail.fl_str_mv |
repositorio@academico.ufs.br |
_version_ |
1802110848686096384 |