Simulation-based verification of false positive data training for tunnel incident detection systems

Kyu Beom Lee; Young Bae Seo; Hyun Seong Bae; Hyu Soung Shin

doi:10.9711/KTAJ.2025.27.5.477

All Issue

2025 Vol.27, Issue 5 Preview Page Next Page

Research Paper

Simulation-based verification of false positive data training for tunnel incident detection systems 터널 돌발상황 감지시스템을 위한 오탐지 학습법의 시뮬레이션 기반 현장적용성 검증

30 September 2025. pp. 477-488

PDF XML

Abstract

The tunnel CCTV-based incident detection system is designed to detect abnormal events in real time using CCTV footage inside tunnels. However, its performance is often degraded due to various environmental factors such as low illumination, perspective distortion, camera angle variation, and dust accumulation. In particular, frequent false positive detections significantly reduce the system’s reliability. To address this issue, previous studies proposed a false positive data training method, in which false positive data generated during actual operation are collected, corrected, and iteratively added to the training dataset. This approach has been shown to enable self-enhancement of object detection performance through repeated retraining. This study extends the method by applying it to the entire tunnel incident detection system and verifies its applicability and effectiveness through a field application simulation. The experiment was conducted using CCTV footage recorded at the XX Tunnel in 2021. After training the initial model, false positive data occurring at each simulation stage were corrected, accumulated, and used for retraining. Furthermore, to mitigate the imbalance of classes related to pedestrian detection, an oversampling technique was introduced. The results of the field application simulation demonstrated that the proposed method significantly reduced the frequency of false positives in normal traffic scenes, particularly suppressing unnecessary pedestrian false alarms. However, in incident scenarios, improvements in detection rates were limited. This limitation is attributed to factors such as dataset imbalance, bounding box instability during inference, and the wide frame sampling interval. In conclusion, this study confirms that the false positive data training method is effective in enhancing the system’s reliability by reducing false alarms in normal situations. Nonetheless, further improvements in incident detection performance will require more balanced datasets and enhanced stability of deep learning-based inference models.

Keywords

Tunnel CCTV

Incident detection

False positive data training method

Oversampling

Field application simulation

터널 돌발상황 감지시스템은 터널 내 CCTV 영상을 통해 돌발상황을 실시간으로 감지하는 시스템으로, 낮은 조도, 원근 왜곡, 카메라 각도 변화, 먼지 축적 등 다양한 환경적 요인으로 인해 돌발상황 감지 성능이 저하되며, 특히 빈번한 오탐지 발생으로 인해 시스템의 신뢰도가 하락하는 문제가 있다. 이러한 문제를 해결하기 위해 기존 연구에서는 오탐지 학습법을 제안한 바 있다. 이 방법은 실제 운영 과정에서 발생한 오탐지 데이터를 수집·정정한 뒤 이를 원래 학습 데이터셋에 반복적으로 추가하여 재학습하는 방식으로, 반복적인 재학습을 통한 객체인식 성능의 자가 향상이 입증되었다. 본 연구는 이를 한 단계 확장하여 터널 돌발상황 감지시스템 전체 프로세스에 오탐지 학습법을 적용하고, 실제 현장 상황을 모사한 시뮬레이션을 통해 적용 가능성과 성능 개선 효과를 검증하는 것을 목표로 한다. 실험은 2021년도 XX터널 CCTV 영상을 기반으로 진행되었으며, 초기 모델 학습 후 각 주기별로 발생한 오탐지 데이터를 정정·누적하여 재학습하는 절차를 모사하였다. 또한 보행자 관련 클래스의 불균형을 보정하기 위해 오버샘플링 기법을 도입하였다. 현장 적용 시뮬레이션 결과, 정상상황 영상에서 오탐지 발생 빈도가 크게 줄었으며, 특히 보행자에 대한 불필요한 오탐지를 효과적으로 억제할 수 있었다. 그러나 돌발상황 영상에서는 정탐지율의 개선이 제한적으로 나타났으며, 이는 데이터셋의 편향, 추론 단계에서의 경계 박스 흔들림 현상, 프레임 샘플링 간격의 한계 등 복합적인 요인에 기인한 것으로 분석되었다. 결론적으로, 본 연구는 오탐지 학습법이 정상상황에서의 오탐지 저감을 통해 시스템 신뢰성을 높이는 데 효과적임을 입증하였으나, 돌발상황 정탐지 성능 향상을 위해서는 데이터 균형 확보와 딥러닝 객체인식 모델의 추론 안정성 개선이 향후 연구 과제로 요구됨을 확인하였다.

키워드

터널 돌발상황 감지시스템

현장적용 시뮬레이션

오탐지 학습법

오버샘플링 기법

돌발상황 감지

References

Alex, B., Zongyuan, G., Lionel, O., Fabio, R., Ben, U. (2016), “Simple online and realtime tracking”, Proceedings of the IEEE International Conference on Image Processing (ICIP), Phoenix, pp. 3464-3468.

Cao, J., Pang, J., Weng, X., Khirodkar, R., Kitani, K. (2023), “Observation-centric sort: Rethinking sort for robust multi-object tracking”, Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Vancouver, pp. 9686-9696.

10.1109/CVPR52729.2023.00934

Cheu, R.L., Ritchie, S.G. (1995), “Automated detection of lane-blocking freeway incidents using artificial neural networks”, Transportation Research Part C: Emerging Technologies, Vol. 3, No. 6, pp. 371-388.

10.1016/0968-090X(95)00016-C

Dendorfer, P., Osep, A., Milan, A., Schindler, K., Cremers, D., Reid, I., Roth, S., Leal-Taixé, L. (2021), “Motchallenge: A benchmark for single-camera multiple target tracking”, International Journal of Computer Vision, Vol. 129, pp. 845-881.

10.1007/s11263-020-01393-0

KTA (Korean Tunnelling and Underground Space Association) (2015), Study on revision of installation and operation guideline for hazard mitigation facilities of road tunnels, Ministry of Land Infrastructure and Transport (MOLIT), Dec., pp. 326.

Lee, K.B., Shin, H.S. (2019), “Effect on self-enhancement of deep-learning inference by repeated training of false detection cases in tunnel accident image detection”, Journal of Korean Tunnelling and Underground Space Association, Vol. 21, No. 3, pp. 419-432.

10.9711/KTAJ.2019.21.3.419

Lee, K.B., Shin, H.S. (2022), “A study for improvement of far-distance performance of a tunnel accident detection system by using an inverse perspective transformation”, Journal of Korean Tunnelling and Underground Space Association, Vol. 24, No. 3, pp. 247-262.

10.9711/KTAJ.2022.24.3.247

Lee, K.B., Shin, H.S. (2024a), “A study on improving self-inference performance through iterative retraining of false positives of deep-learning object detection in tunnels”, Journal of Korean Tunnelling and Underground Space Association, Vol. 26, No. 2, pp. 129-152.

10.9711/KTAJ.2024.26.2.129

Lee, K.B., Shin, H.S. (2024b), “Development of an automatic region of interest setting algorithm based on tunnel CCTV”, Journal of Korean Tunnelling and Underground Space Association, Vol. 26, No. 6, pp. 563-582.

10.9711/KTAJ.2024.26.6.563

Lee, K.B., Shin, H.S., Kim, D.G. (2018), “Development of a deep-learning based automatic tracking of moving vehicles and incident detection processes on tunnels”, Journal of Korean Tunnelling and Underground Space Association, Vol. 20, No. 6, pp. 1161-1175.

10.9711/KTAJ.2018.20.6.1161

Mohammed, R., Rawashdeh, J., Abdullah, M. (2020), “Machine learning with oversampling and undersampling techniques: Overview study and experimental results”, Proceedings of the 11th International Conference on Information and Communication Systems (ICICS), Irbid, pp. 243-248.

10.1109/ICICS49469.2020.239556

MOLIT (2021), Guideline of installation and management of disaster prevention facilities on road tunnels, Ministry of Land, Infrastructure and Transport.

Prevedouros, P.D., Ji, X., Papandreou, K., Kopelias, P., Vegiri, V. (2006), “Video incident detection tests in freeway tunnels”, Transportation Research Record, Vol. 1959, No. 1, pp. 130-139.

10.1177/0361198106195900115

Shin, H.S., Lee, K.B., Yim, M.J., Kim, D.G. (2017), “Development of a deep-learning based tunnel incident detection system on CCTVs”, Journal of Korean Tunnelling and Underground Space Association, Vol. 19, No.6, pp. 915-936.

10.9711/KTAJ.2017.19.1.095

Wang, C.Y., Bochkovskiy, A., Liao, H.Y.M. (2023), “YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors”, Proceedings of the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Vancouver, pp. 7464-7475.

10.1109/CVPR52729.2023.00721

Information

Publisher :Korean Tunneling and Underground Space Association
Publisher(Ko) :한국터널지하공간학회
Journal Title :Journal of Korean Tunnelling and Underground Space Association
Journal Title(Ko) :한국터널지하공간학회 논문집
Volume : 27
No :5
Pages :477-488
Received Date : 2025-09-08
Revised Date : 2025-09-22
Accepted Date : 2025-09-22
DOI :https://doi.org/10.9711/KTAJ.2025.27.5.477

Journal of Korean Tunnelling and Underground Space Association ISSN:2233-8292(Print) 2287-4747(Online) 한국터널지하공간학회 논문집

All Issue