Abstract

Person re-identification is the process of finding people across different cameras. In this process, focus often lies in developing strong feature descriptors or a robust metric learning algorithm. While the two aspects are the most important steps in order to secure a high performance, a less explored aspect is late fusion of complementary features. For this purpose, this study proposes a late fusing scheme that, based on an experimental analysis, combines three systems that focus on extracting features and provide supervised learning on different abstraction levels. To analyse the behaviour of the proposed system, both rank aggregation and score-level fusion are applied. The authors’ proposed fusion scheme increases results on both small and large datasets. Experimental results on VIPeR show accuracies 5.43% higher than related systems, while results on PRID450S and CUHK01 increase state-of-the-art results by 10.94 and 14.84%, respectively. Furthermore, a cross-dataset test shows an increased rank-1 accuracy of 28.26% when training on CUHK02 and testing on VIPeR. Finally, an analysis of the late fusion shows aggregation to be better when individual results are unequally distributed within top-10 while score-level fusion provides better results when two individual results lie within top-5 while the last lies outside top-10.
Original languageEnglish
JournalIET Biometrics
Volume7
Issue number2
Pages (from-to)125-135
Number of pages22
ISSN2047-4946
DOIs
Publication statusPublished - 2018

Fingerprint

Fusion reactions
Agglomeration
Supervised learning
Learning algorithms
Cameras
Testing

Keywords

    Cite this

    @article{c6f3915c5b5247b897277affb45bb1b5,
    title = "Enhancing Person Re-identification by Late Fusion of Low-, Mid-, and High-Level Features",
    abstract = "Person re-identification is the process of finding people across different cameras. In this process, focus often lies in developing strong feature descriptors or a robust metric learning algorithm. While the two aspects are the most important steps in order to secure a high performance, a less explored aspect is late fusion of complementary features. For this purpose, this study proposes a late fusing scheme that, based on an experimental analysis, combines three systems that focus on extracting features and provide supervised learning on different abstraction levels. To analyse the behaviour of the proposed system, both rank aggregation and score-level fusion are applied. The authors’ proposed fusion scheme increases results on both small and large datasets. Experimental results on VIPeR show accuracies 5.43{\%} higher than related systems, while results on PRID450S and CUHK01 increase state-of-the-art results by 10.94 and 14.84{\%}, respectively. Furthermore, a cross-dataset test shows an increased rank-1 accuracy of 28.26{\%} when training on CUHK02 and testing on VIPeR. Finally, an analysis of the late fusion shows aggregation to be better when individual results are unequally distributed within top-10 while score-level fusion provides better results when two individual results lie within top-5 while the last lies outside top-10.",
    keywords = "Person re-identification, Maskin l{\ae}ring, Neural netv{\ae}rk",
    author = "Lejb{\o}lle, {Aske Rasch} and Kamal Nasrollahi and Moeslund, {Thomas B.}",
    year = "2018",
    doi = "10.1049/iet-bmt.2016.0200",
    language = "English",
    volume = "7",
    pages = "125--135",
    journal = "IET Biometrics",
    issn = "2047-4938",
    publisher = "TheInstitution of Engineering and Technology",
    number = "2",

    }

    Enhancing Person Re-identification by Late Fusion of Low-, Mid-, and High-Level Features. / Lejbølle, Aske Rasch; Nasrollahi, Kamal; Moeslund, Thomas B.

    In: IET Biometrics, Vol. 7, No. 2, 2018, p. 125-135.

    Research output: Contribution to journalJournal articleResearchpeer-review

    TY - JOUR

    T1 - Enhancing Person Re-identification by Late Fusion of Low-, Mid-, and High-Level Features

    AU - Lejbølle, Aske Rasch

    AU - Nasrollahi, Kamal

    AU - Moeslund, Thomas B.

    PY - 2018

    Y1 - 2018

    N2 - Person re-identification is the process of finding people across different cameras. In this process, focus often lies in developing strong feature descriptors or a robust metric learning algorithm. While the two aspects are the most important steps in order to secure a high performance, a less explored aspect is late fusion of complementary features. For this purpose, this study proposes a late fusing scheme that, based on an experimental analysis, combines three systems that focus on extracting features and provide supervised learning on different abstraction levels. To analyse the behaviour of the proposed system, both rank aggregation and score-level fusion are applied. The authors’ proposed fusion scheme increases results on both small and large datasets. Experimental results on VIPeR show accuracies 5.43% higher than related systems, while results on PRID450S and CUHK01 increase state-of-the-art results by 10.94 and 14.84%, respectively. Furthermore, a cross-dataset test shows an increased rank-1 accuracy of 28.26% when training on CUHK02 and testing on VIPeR. Finally, an analysis of the late fusion shows aggregation to be better when individual results are unequally distributed within top-10 while score-level fusion provides better results when two individual results lie within top-5 while the last lies outside top-10.

    AB - Person re-identification is the process of finding people across different cameras. In this process, focus often lies in developing strong feature descriptors or a robust metric learning algorithm. While the two aspects are the most important steps in order to secure a high performance, a less explored aspect is late fusion of complementary features. For this purpose, this study proposes a late fusing scheme that, based on an experimental analysis, combines three systems that focus on extracting features and provide supervised learning on different abstraction levels. To analyse the behaviour of the proposed system, both rank aggregation and score-level fusion are applied. The authors’ proposed fusion scheme increases results on both small and large datasets. Experimental results on VIPeR show accuracies 5.43% higher than related systems, while results on PRID450S and CUHK01 increase state-of-the-art results by 10.94 and 14.84%, respectively. Furthermore, a cross-dataset test shows an increased rank-1 accuracy of 28.26% when training on CUHK02 and testing on VIPeR. Finally, an analysis of the late fusion shows aggregation to be better when individual results are unequally distributed within top-10 while score-level fusion provides better results when two individual results lie within top-5 while the last lies outside top-10.

    KW - Person re-identification

    KW - Maskin læring

    KW - Neural netværk

    U2 - 10.1049/iet-bmt.2016.0200

    DO - 10.1049/iet-bmt.2016.0200

    M3 - Journal article

    VL - 7

    SP - 125

    EP - 135

    JO - IET Biometrics

    JF - IET Biometrics

    SN - 2047-4938

    IS - 2

    ER -