DOI QR코드

DOI QR Code

Effect on self-enhancement of deep-learning inference by repeated training of false detection cases in tunnel accident image detection

터널 내 돌발상황 오탐지 영상의 반복 학습을 통한 딥러닝 추론 성능의 자가 성장 효과

  • Lee, Kyu Beom (Department of Future Technology and Convergence Research, Korea Institute of Civil Engineering and Building Technology, Smart City and University of Science & Technology) ;
  • Shin, Hyu Soung (Department of Future Technology and Convergence Research, Korea Institute of Civil Engineering and Building Technology)
  • 이규범 (한국건설기술연구원 미래융합연구본부, 과학기술연합대학원대학교(UST) 스마트도시건설융합) ;
  • 신휴성 (한국건설기술연구원 미래융합연구본부)
  • Received : 2019.04.01
  • Accepted : 2019.05.02
  • Published : 2019.05.31

Abstract

Most of deep learning model training was proceeded by supervised learning, which is to train labeling data composed by inputs and corresponding outputs. Labeling data was directly generated manually, so labeling accuracy of data is relatively high. However, it requires heavy efforts in securing data because of cost and time. Additionally, the main goal of supervised learning is to improve detection performance for 'True Positive' data but not to reduce occurrence of 'False Positive' data. In this paper, the occurrence of unpredictable 'False Positive' appears by trained modes with labeling data and 'True Positive' data in monitoring of deep learning-based CCTV accident detection system, which is under operation at a tunnel monitoring center. Those types of 'False Positive' to 'fire' or 'person' objects were frequently taking place for lights of working vehicle, reflecting sunlight at tunnel entrance, long black feature which occurs to the part of lane or car, etc. To solve this problem, a deep learning model was developed by simultaneously training the 'False Positive' data generated in the field and the labeling data. As a result, in comparison with the model that was trained only by the existing labeling data, the re-inference performance with respect to the labeling data was improved. In addition, re-inference of the 'False Positive' data shows that the number of 'False Positive' for the persons were more reduced in case of training model including many 'False Positive' data. By training of the 'False Positive' data, the capability of field application of the deep learning model was improved automatically.

대부분 딥러닝 모델의 학습은 입력값과 입력값에 따른 출력값이 포함된 레이블링 데이터(labeling data)를 학습하는 지도 학습(supervised learning)으로 진행된다. 레이블링 데이터는 인간이 직접 제작하므로 데이터의 정확도가 높다는 장점이 있지만 비용과 시간의 문제로 인해 데이터의 확보에 많은 노력이 소요된다. 그리고 지도 학습의 목표는 정탐지 데이터(true positive data)의 인식 성능 향상에 초점이 맞추어져 있으며, 오탐지 데이터(false positive data)의 발생에 대한 대처는 미흡한 실정이다. 본 논문은 터널 관제센터에 투입된 딥러닝 모델 기반 영상유고 시스템의 모니터링을 통해 정탐지와 레이블링 데이터의 학습으로 예측하기 힘든 오탐지의 발생을 확인하였다. 오탐지의 유형은 작업차량의 경광등, 터널 입구부에서 반사되는 햇빛, 차선과 차량의 일부에서 발생하는 길쭉한 검은 음영 등이 화재와 보행자로 오탐지되고 있었다. 이러한 문제를 해결하기 위해 현장에서 발생한 오탐지 데이터와 레이블링 데이터를 동시에 학습하여 딥러닝 모델을 개발하였으며, 그 결과 기존 레이블링 데이터만 학습한 모델과 비교하면 레이블링 데이터에 대한 재추론 성능이 향상됨을 알 수 있었다. 그리고 오탐지 데이터에 대한 재추론을 한 결과 오탐지 데이터를 많이 포함하여 학습한 모델일 경우 보행자의 오탐지 개수가 훨씬 줄었으며, 오탐지 데이터의 학습을 통해 딥러닝 모델의 현장 적용성을 향상시킬 수 있었다.

Keywords

TNTNB3_2019_v21n3_419_f0001.png 이미지

Fig. 1. ‘False Positive’ report for person and fire objects during 55 days after installing deep learning-based CCTV accident detection system

TNTNB3_2019_v21n3_419_f0002.png 이미지

Fig. 2. ‘False Positive’ samples occurred by deep learning-based accident detection system in tunnel monitoring

TNTNB3_2019_v21n3_419_f0003.png 이미지

Fig. 3. Similarity between ‘True Positive’ and ‘False Positive’ samples for fire and person objects

TNTNB3_2019_v21n3_419_f0004.png 이미지

Fig. 4. Deep learning training process including False Positive data

TNTNB3_2019_v21n3_419_f0005.png 이미지

Fig. 5. Test results of deep learning models

TNTNB3_2019_v21n3_419_f0006.png 이미지

Fig. 6. ‘False Positive’ report for fire objects during 97 days after installing deep learning-based CCTV accident detection system

Table 1. 4 kinds of prediction result for testing data

TNTNB3_2019_v21n3_419_t0001.png 이미지

Table 2. Composition of labeling data and ‘False Positive’ data

TNTNB3_2019_v21n3_419_t0002.png 이미지

Table 3. Training dataset models

TNTNB3_2019_v21n3_419_t0003.png 이미지

Table 4. Training condition of a deep learning algorithm

TNTNB3_2019_v21n3_419_t0004.png 이미지

Table 5. The number of ‘False Positives’ resulted in site from trained models with ‘False Positive’ data

TNTNB3_2019_v21n3_419_t0005.png 이미지

References

  1. Davis, J., Goadrich, M. (2006), "The relationship between Precision-Recall and ROC curves", Proceedings of the 23rd International Conference on Machine Learning, pp. 233-240.
  2. Everingham, M., Gool, L.V., Williams, C.K.I., Winn, J., Zisserman, A. (2010), "The pascal visual object classes (VOC) challenge", International Journal of Computer Vision, Vol. 88, No. 2, pp. 303-338. https://doi.org/10.1007/s11263-009-0275-4
  3. Kim, D.G., Shin, Y.W., Shin, Y.S. (2012), "Section enlargement by reinforcement of shotcrete lining on the side wall of operating road tunnel", Journal of Korean Tunnelling and Underground Space Association, Vol. 14, No. 6, pp. 637-652. https://doi.org/10.9711/KTAJ.2012.14.6.637
  4. Kim, T.B. (2016), "The national highway, expressway tunnel video incident detection system performance analysis and reflect attributes for double deck tunnel in great depth underground space", Journal of Korea Institute of Information and Communication Engineering, Vol. 20, No. 7, pp. 1325-1334. https://doi.org/10.6109/jkiice.2016.20.7.1325
  5. LeCun, Y., Yoshua, B., Geoffrey, H. (2015), "Deep learning", Nature, Vol. 521, pp. 436-444. https://doi.org/10.1038/nature14539
  6. Lee, K.B., Shin, H.S., Kim, D.K. (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. https://doi.org/10.9711/KTAJ.2018.20.6.1161
  7. Lin, T.Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollar, P., Zitnick, C.L. (2014), "Microsoft coco: Common objects in context", European Conference on Computer Vision. Springer, Cham, pp. 740-755.
  8. Ministry of Land, Infrastructure and Transport (MOLIT) (2016), Guideline of installation and management of disaster prevention facilities on road tunnels.
  9. National Committee for Land and Transport (2016), "Tunnel accidents increase, but tunnel incident automatic detection system often fails in operation", Press Release from parliamentary inspection of MOLIT. Congressman Yoon Hoo-Dyuk.
  10. Ren, S., He, K., Girshick, R., Sun, J. (2015), "Faster R-CNN: Towards real-time object detection with region proposal networks", Proceedings of the Advances in Neural Information Processing Systems, pp. 91-99.
  11. Shin, H.S., Lee, K.B., Yim, M.J., Kim, D.K. (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. https://doi.org/10.9711/KTAJ.2017.19.6.915
  12. Zhu, M. (2004), "Recall, precision and average precision", Department of Statistics and Actuarial Science, University of Waterloo, Waterloo 2: 30.