Publication detail

Graph@FIT Submission to the NVIDIA AI City Challenge 2018

SOCHOR, J. ŠPAŇHEL, J. JURÁNEK, R. DOBEŠ, P. HEROUT, A.

Original Title

Graph@FIT Submission to the NVIDIA AI City Challenge 2018

English Title

Graph@FIT Submission to the NVIDIA AI City Challenge 2018

Type

conference paper

Language

en

Original Abstract

In our submission to the NVIDIA AI City Challenge, we address speed measurement of vehicles and vehicle re-identification. For both these tasks, we use a calibration method based on extracted vanishing points. We detect and track vehicles by a CNN-based detector and we construct 3D bounding boxes for all vehicles. For the speed measurement task, we estimate the speed from the movement of the bounding box in the 3D space using the calibration. Our approach to vehicle re-identification is based on extraction of visual features from "unpacked" images of the vehicles. The features are aggregated in temporal domain to obtain a single feature descriptor for the whole track. Furthermore, we utilize a validation network to improve the re-identification accuracy.

English abstract

In our submission to the NVIDIA AI City Challenge, we address speed measurement of vehicles and vehicle re-identification. For both these tasks, we use a calibration method based on extracted vanishing points. We detect and track vehicles by a CNN-based detector and we construct 3D bounding boxes for all vehicles. For the speed measurement task, we estimate the speed from the movement of the bounding box in the 3D space using the calibration. Our approach to vehicle re-identification is based on extraction of visual features from "unpacked" images of the vehicles. The features are aggregated in temporal domain to obtain a single feature descriptor for the whole track. Furthermore, we utilize a validation network to improve the re-identification accuracy.

Keywords

vehicle speed measurement, vehicle re-identification, challenge, camera calibration

Released

09.04.2018

Publisher

IEEE Computer Society

Location

Salt Lake City

ISBN

978-1-5386-6100-0

Book

NVIDIA AI City Challenge 2018 (CVPRW)

Edition

NEUVEDEN

Edition number

NEUVEDEN

Pages from

77

Pages to

84

Pages count

8

URL

Documents

BibTex


@inproceedings{BUT155029,
  author="Jakub {Sochor} and Jakub {Špaňhel} and Roman {Juránek} and Petr {Dobeš} and Adam {Herout}",
  title="Graph@FIT Submission to the NVIDIA AI City Challenge 2018",
  annote="In our submission to the NVIDIA AI City Challenge, we address speed measurement
of vehicles and vehicle re-identification. For both these tasks, we use
a calibration method based on extracted vanishing points. We detect and track
vehicles by a CNN-based detector and we construct 3D bounding boxes for all
vehicles. For the speed measurement task, we estimate the speed from the movement
of the bounding box in the 3D space using the calibration. Our approach to
vehicle re-identification is based on extraction of visual features from
"unpacked" images of the vehicles. The features are aggregated in temporal domain
to obtain a single feature descriptor for the whole track. Furthermore, we
utilize a validation network to improve the re-identification accuracy.",
  address="IEEE Computer Society",
  booktitle="NVIDIA AI City Challenge 2018 (CVPRW)",
  chapter="155029",
  doi="10.1109/CVPRW.2018.00018",
  edition="NEUVEDEN",
  howpublished="online",
  institution="IEEE Computer Society",
  year="2018",
  month="april",
  pages="77--84",
  publisher="IEEE Computer Society",
  type="conference paper"
}