Deep 2.5D Vehicle Classification with Sparse SfM Depth Prior for Automated Toll Systems

Georg Waltner, Michael Maurer, Thomas Holzmann, Patrick Ruprecht, Michael Opitz, Horst Possegger, Friedrich Fraundorfer, Horst Bischof

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Research › peer-review

Abstract

Automated toll systems rely on proper classification of the passing vehicles. This is especially difficult when the images used for classification only cover parts of the vehicle. To obtain information about the whole vehicle. we reconstruct the vehicle as 3D object and exploit this additional information within a Convolutional Neural Network (CNN). However, when using deep networks for 3D object classification, large amounts of dense 3D models are required for good accuracy, which are often neither available nor feasible to process due to memory requirements. Therefore, in our method we reproject the 3D object onto the image plane using the reconstructed points, lines or both. We utilize this sparse depth prior within an auxiliary network branch that acts as a regularizer during training. We show that this auxiliary regularizer helps to improve accuracy compared to 2D classification on a real-world dataset. Furthermore due to the design of the network, at test time only the 2D camera images are required for classification which enables the usage in portable computer vision systems.
Original languageEnglish
Title of host publication2018 IEEE Intelligent Transportation Systems Conference, ITSC 2018
PublisherInstitute of Electrical and Electronics Engineers
Pages3212-3217
Number of pages6
Volume2018-November
ISBN (Electronic)9781728103235
DOIs
Publication statusPublished - 7 Dec 2018
Event21st IEEE International Conference on Intelligent Transportation Systems - Maui, United States
Duration: 4 Nov 2018 → 7 Nov 2018

Conference

Conference21st IEEE International Conference on Intelligent Transportation Systems
Abbreviated titleITSC
CountryUnited States
CityMaui
Period4/11/18 → 7/11/18

Fingerprint

Computer vision
Cameras
Neural networks
Data storage equipment

Cite this

Waltner, G., Maurer, M., Holzmann, T., Ruprecht, P., Opitz, M., Possegger, H., ... Bischof, H. (2018). Deep 2.5D Vehicle Classification with Sparse SfM Depth Prior for Automated Toll Systems. In 2018 IEEE Intelligent Transportation Systems Conference, ITSC 2018 (Vol. 2018-November, pp. 3212-3217). [8569670] Institute of Electrical and Electronics Engineers. https://doi.org/10.1109/ITSC.2018.8569670

Deep 2.5D Vehicle Classification with Sparse SfM Depth Prior for Automated Toll Systems. / Waltner, Georg; Maurer, Michael; Holzmann, Thomas; Ruprecht, Patrick; Opitz, Michael; Possegger, Horst; Fraundorfer, Friedrich; Bischof, Horst.

2018 IEEE Intelligent Transportation Systems Conference, ITSC 2018. Vol. 2018-November Institute of Electrical and Electronics Engineers, 2018. p. 3212-3217 8569670.

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Research › peer-review

Waltner, G, Maurer, M, Holzmann, T, Ruprecht, P, Opitz, M, Possegger, H, Fraundorfer, F & Bischof, H 2018, Deep 2.5D Vehicle Classification with Sparse SfM Depth Prior for Automated Toll Systems. in 2018 IEEE Intelligent Transportation Systems Conference, ITSC 2018. vol. 2018-November, 8569670, Institute of Electrical and Electronics Engineers, pp. 3212-3217, 21st IEEE International Conference on Intelligent Transportation Systems, Maui, United States, 4/11/18. https://doi.org/10.1109/ITSC.2018.8569670
Waltner G, Maurer M, Holzmann T, Ruprecht P, Opitz M, Possegger H et al. Deep 2.5D Vehicle Classification with Sparse SfM Depth Prior for Automated Toll Systems. In 2018 IEEE Intelligent Transportation Systems Conference, ITSC 2018. Vol. 2018-November. Institute of Electrical and Electronics Engineers. 2018. p. 3212-3217. 8569670 https://doi.org/10.1109/ITSC.2018.8569670
Waltner, Georg ; Maurer, Michael ; Holzmann, Thomas ; Ruprecht, Patrick ; Opitz, Michael ; Possegger, Horst ; Fraundorfer, Friedrich ; Bischof, Horst. / Deep 2.5D Vehicle Classification with Sparse SfM Depth Prior for Automated Toll Systems. 2018 IEEE Intelligent Transportation Systems Conference, ITSC 2018. Vol. 2018-November Institute of Electrical and Electronics Engineers, 2018. pp. 3212-3217
@inproceedings{10b52b92a9384fd79ad5b0fe8392fd97,
title = "Deep 2.5D Vehicle Classification with Sparse SfM Depth Prior for Automated Toll Systems",
abstract = "Automated toll systems rely on proper classification of the passing vehicles. This is especially difficult when the images used for classification only cover parts of the vehicle. To obtain information about the whole vehicle. we reconstruct the vehicle as 3D object and exploit this additional information within a Convolutional Neural Network (CNN). However, when using deep networks for 3D object classification, large amounts of dense 3D models are required for good accuracy, which are often neither available nor feasible to process due to memory requirements. Therefore, in our method we reproject the 3D object onto the image plane using the reconstructed points, lines or both. We utilize this sparse depth prior within an auxiliary network branch that acts as a regularizer during training. We show that this auxiliary regularizer helps to improve accuracy compared to 2D classification on a real-world dataset. Furthermore due to the design of the network, at test time only the 2D camera images are required for classification which enables the usage in portable computer vision systems.",
author = "Georg Waltner and Michael Maurer and Thomas Holzmann and Patrick Ruprecht and Michael Opitz and Horst Possegger and Friedrich Fraundorfer and Horst Bischof",
year = "2018",
month = "12",
day = "7",
doi = "10.1109/ITSC.2018.8569670",
language = "English",
volume = "2018-November",
pages = "3212--3217",
booktitle = "2018 IEEE Intelligent Transportation Systems Conference, ITSC 2018",
publisher = "Institute of Electrical and Electronics Engineers",
address = "United States",

}

TY - GEN

T1 - Deep 2.5D Vehicle Classification with Sparse SfM Depth Prior for Automated Toll Systems

AU - Waltner, Georg

AU - Maurer, Michael

AU - Holzmann, Thomas

AU - Ruprecht, Patrick

AU - Opitz, Michael

AU - Possegger, Horst

AU - Fraundorfer, Friedrich

AU - Bischof, Horst

PY - 2018/12/7

Y1 - 2018/12/7

N2 - Automated toll systems rely on proper classification of the passing vehicles. This is especially difficult when the images used for classification only cover parts of the vehicle. To obtain information about the whole vehicle. we reconstruct the vehicle as 3D object and exploit this additional information within a Convolutional Neural Network (CNN). However, when using deep networks for 3D object classification, large amounts of dense 3D models are required for good accuracy, which are often neither available nor feasible to process due to memory requirements. Therefore, in our method we reproject the 3D object onto the image plane using the reconstructed points, lines or both. We utilize this sparse depth prior within an auxiliary network branch that acts as a regularizer during training. We show that this auxiliary regularizer helps to improve accuracy compared to 2D classification on a real-world dataset. Furthermore due to the design of the network, at test time only the 2D camera images are required for classification which enables the usage in portable computer vision systems.

AB - Automated toll systems rely on proper classification of the passing vehicles. This is especially difficult when the images used for classification only cover parts of the vehicle. To obtain information about the whole vehicle. we reconstruct the vehicle as 3D object and exploit this additional information within a Convolutional Neural Network (CNN). However, when using deep networks for 3D object classification, large amounts of dense 3D models are required for good accuracy, which are often neither available nor feasible to process due to memory requirements. Therefore, in our method we reproject the 3D object onto the image plane using the reconstructed points, lines or both. We utilize this sparse depth prior within an auxiliary network branch that acts as a regularizer during training. We show that this auxiliary regularizer helps to improve accuracy compared to 2D classification on a real-world dataset. Furthermore due to the design of the network, at test time only the 2D camera images are required for classification which enables the usage in portable computer vision systems.

UR - https://arxiv.org/abs/1805.03511

UR - http://www.scopus.com/inward/record.url?scp=85060441996&partnerID=8YFLogxK

U2 - 10.1109/ITSC.2018.8569670

DO - 10.1109/ITSC.2018.8569670

M3 - Conference contribution

VL - 2018-November

SP - 3212

EP - 3217

BT - 2018 IEEE Intelligent Transportation Systems Conference, ITSC 2018

PB - Institute of Electrical and Electronics Engineers

ER -

Access to Document