An Embedded Deep Learning-Based Package for Traffic Law Enforcement

Abbas Omidi, Amirhossein Heydarian, Aida Mohammadshahi, Behnam Asghari Beirami, Farzan Haddadi; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2021, pp. 262-271

Abstract


Crossing Heavy Good Vehicles (HGVs) from the overtaking lane in highways is not only a traffic violation but may also cause severe casualties in case of an accident happening in such velocities. Currently, the only way to prevent this violation is to identify the violating vehicles by the highway police, so in this paper, a violation detection system using an embedded camera is introduced using algorithms based on deep learning and image processing techniques. The embedded system benefits of a multi-stage deep system based on the YOLO network, which consists of four stages of cascaded detection, including overtaking lane detection, HGV detection, license plate detection, and character recognition. In this research, the developed deep learning models, after some initial training, are fine-tuned on a local Persian dataset collected with distributed cameras. The accuracy obtained on the test dataset of each of the four separate stages was above 85% and the results show the efficiency of the proposed smart system with 70% accuracy in the union of all stages. All data including local datasets, implementations, codes, and results are available on the project's GitHub (https://github.com/NEFTeam/Traffic-Law-Enforcement).

Related Material


[pdf]
[bibtex]
@InProceedings{Omidi_2021_ICCV, author = {Omidi, Abbas and Heydarian, Amirhossein and Mohammadshahi, Aida and Beirami, Behnam Asghari and Haddadi, Farzan}, title = {An Embedded Deep Learning-Based Package for Traffic Law Enforcement}, booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops}, month = {October}, year = {2021}, pages = {262-271} }