Autonomous Reckless Driving Detection Using Deep Learning on Embedded GPUs

Taewook Heo, Woojin Nam, Jeongyeup Paek, Jeonggil Ko

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Reckless driving is dangerous, and must be monitored, detected, and law-enforced to assure road safety. For this purpose, this work presents an embedded system for monitoring and detecting reckless driving activities on the road autonomously in real-time. Using an embedded GPU (eGPU) platform, a camera, and a combination of light-weight deep learning models, we design a system that can identify abnormal vehicle motions on the road. Our system analyzes discrete per-frame images from vehicle detection algorithms, and creates a continuous trace of a vehicle's motion trajectory. While doing so, a virtual grid is generated on the road to obtain positions of vehicles with less overhead and accurately track a vehicle's movement even with low frame rate (5fps) videos. Vehicle's motion trajectory is then compared against the surrounding to identify abnormal behavior through driving activity classification, which can be provided to law enforcement personnel for final validation. The key challenge is the fundamental resource constraints of embedded platforms, and we design algorithms to overcome their limitations. Evaluation results show that our scheme can wellextract the horizontal and vertical movements of a vehicle (100% recall and 67% precision) and show the potential for truly autonomous reckless driving activity detection systems.

Original languageEnglish
Title of host publicationProceedings - 2020 IEEE 17th International Conference on Mobile Ad Hoc and Smart Systems, MASS 2020
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages464-472
Number of pages9
ISBN (Electronic)9781728198668
DOIs
Publication statusPublished - 2020 Dec
Event17th IEEE International Conference on Mobile Ad Hoc and Smart Systems, MASS 2020 - Virtual, Delhi, India
Duration: 2020 Dec 102020 Dec 13

Publication series

NameProceedings - 2020 IEEE 17th International Conference on Mobile Ad Hoc and Smart Systems, MASS 2020

Conference

Conference17th IEEE International Conference on Mobile Ad Hoc and Smart Systems, MASS 2020
Country/TerritoryIndia
CityVirtual, Delhi
Period20/12/1020/12/13

Bibliographical note

Funding Information:
This work was supported by the NRF Grant funded by MSIP (Project No. 2015R1A5A1037668) and KETEP and MOTIE of the Republic of Korea (No. 20182010106460).

Publisher Copyright:
© 2020 IEEE.

All Science Journal Classification (ASJC) codes

  • Artificial Intelligence
  • Computer Networks and Communications
  • Computer Science Applications
  • Hardware and Architecture
  • Information Systems and Management
  • Safety, Risk, Reliability and Quality

Fingerprint

Dive into the research topics of 'Autonomous Reckless Driving Detection Using Deep Learning on Embedded GPUs'. Together they form a unique fingerprint.

Cite this