DOI QR코드

DOI QR Code

Research on the Detection Framework for Dangerous Riding Electric Scooters

위험 주행 전동 킥보드 감지 프레임워크 연구

  • 황서빈 (전남대학교 인공지능융합학과) ;
  • 조영준 (전남대학교 인공지능융합학과)
  • Received : 2024.04.26
  • Accepted : 2024.06.25
  • Published : 2024.10.31

Abstract

E-scooters are eco-friendly and convenient, making their rental services highly popular. However, simultaneously, due to issues such as a surge in user numbers and a lack of user awareness about relevant traffic laws, related accident rates have increased tenfold in the last five years. As a result, dangerous riding of e-scooters is being presented as a new social issue. This study proposes a framework for detecting dangerously operating e-scooters in a fixed single-camera environment, which is cost-effective and conducive to accident detection. The proposed method uses object detection and tracking technology to identify people and e-scooters, simultaneously detecting multiple riders, helmet non-use, and sidewalk riding. For validation, it achieved excellent dangerous behavior detection performance in 17 diverse scenarios directly generated. Furthermore, compared to existing methods, it could detect more dangerous riding behaviors and provided detailed information, such as separately mapping dangerous riding results for each subject during multiple-rider situations. These results are expected to play a crucial role in enhancing urban traffic safety.

전동 킥보드는 친환경적이고 편리해 대여서비스가 큰 인기를 끌고 있다. 그러나 동시에 사용자 수 급증과 관련 교통법규에 대한 사용자 인식 부족 등의 문제로 최근 5년 간 관련 사고율이 10배 증가해 전동 킥보드 위험 주행이 새로운 사회 안건으로 제시되고 있다. 본 연구는 저렴한 비용으로 사고 감지에 용이한 고정형 단일 카메라 환경에서 위험 주행 중인 전동 킥보드를 감지하는 프레임워크를 제안한다. 제안 방법은 객체 검출과 추적 기술을 사용해 사람과 킥보드를 식별하고, 다인 탑승, 헬멧 미착용, 인도주행 등을 동시에 감지한다. 검증을 위해 직접 생성한 17개의 다양한 시나리오에서 우수한 위험 행위 감지 성능을 달성했다. 또한, 기존 방법들과 비교하여 더 많은 위험 주행 행위를 감지할 수 있었으며, 특히 다인 탑승 시 각 주체별로 위험 주행결과를 별도로 매핑하는 등 상세한 정보를 제공했다. 이러한 결과는 도시 교통 안전을 증진하는 데 중요한 역할을 할 것으로 기대된다.

Keywords

Acknowledgement

이 논문은 정부(과학기술정보통신부)의 재원으로 한국연구재단의 지원을 받아 수행된 연구(No. RS-2022-00165919 ) 이며 과학기술정보통신부 및 정보통신기획평가원의 인공지능융합혁신인재양성사업 연구 결과로 수행되었음(IITP-2023-RS-2023-00256629) 또한 현대차 정몽구 재단 장학생으로서 지원을 받아 수행된 연구임

References

  1. Electric scooters cutting pollution... Safety rules followed, usage up! Benefits rise with new tech! (2022), https://www.dailyt.co.kr/newsView/dlt202208120001, (accessed Oct., 11, 2023).
  2. Electric scooter accidents increase tenfold in five years... Liability compensation system full of loopholes.(2023), https://biz.sbs.co.kr/article/20000119334, (accessed Oct., 11, 2023).
  3. Why are accidents recurring despite regulations on electric scooters? (2022), https://news.kbs.co.kr/news/pc/view/view.do?ncd=5472567, (accessed Oct., 11, 2023).
  4. Neuron Mobility will rectify issues with electric kickboards such as riding on sidewalks, carrying two passengers, and illegal parking(2021), https://zdnet.co.kr/view/?no=20211109225442, (accessed Oct., 11, 2023).
  5. UTech has developed an electric kickboard equipped with a LiDAR sensor 'driving safety device'(2020), https://www.enewstoday.co.kr/news/articleView.html?idxno=1434711, (accessed Oct., 11, 2023).
  6. Siebert, Felix W., et al. "Automated detection of e-scooter helmet use with deep learning.", 2022.
  7. Kim Da-un, Jeong Jin-woo, "Detection of Electric Kickboard Helmets in Nighttime Driving Environments Based on Deep Learning," Journal of the Korean Institute of Electrical Engineers, Vol. 71, No. 10, pp. 1411-1419, 2022. https://doi.org/10.5370/KIEE.2022.71.10.1411
  8. Choi Ye-won, Kim Chan-wook, Lee Hyun-sung. Deep Learning System for Identifying Illegal Acts of Electric Kickboards. 2nd Korean Artificial Intelligence Conference, 2021.
  9. Policy Briefing of the Republic of Korea, Essential Safety Rules for Driving Electric Kickboards, https://www.korea.kr/news/healthView.do?newsId=148909494(2022), (accessed Oct., 11, 2023).
  10. The original electric kickboard was the 'Autoped' from the U.S. in 1915...it failed to become popular due to its high price and lack of a seat(2019), https://www.sedaily.com/NewsView/1VPHCS6LED (accessed Oct., 11, 2023).
  11. 99% of those not wearing motorcycle helmets suffer serious injuries(2012), https://news.kbs.co.kr/news/pc/view/view.do?ncd=2494386 (accessed Oct., 11, 2023).
  12. In electric kickboard accidents, 85 out of 100 people were not wearing a helmet(2022), https://m.health.chosun.com/svc/news_view.html?contid=2022061300832 (accessed Oct., 11, 2023).
  13. Zhou, Yongheng, et al. "Helmet Detection Algorithm Based on Single Pixel Zoom." Journal of Physics: Conference Series, Vol. 1682, No. 1, 2020.
  14. AI city challenge(2023), https://www.aicitychallenge.org (accessed Mar., 15, 2023),
  15. Bike helmet detection(2021), https://universe.roboflow.com/bike-helmets/bike-helmet-detection-2vdjo (accessed Mar., 15, 2023).
  16. Helmet detection(2020), https://www.kaggle.com/datasets/andrewmvd/helmet-detection (accessed Mar, 15, 2023).
  17. YOLOv8(2023), https://github.com/ultralytics (accessed Mar., 30, 2023).
  18. Images of sidewalk riding(2019), https://aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=realm&dataSetSn=189 (accessed Mar., 30, 2023).