Journal of the Korean Institute of Gas (한국가스학회지)

The Korean Institute of GAS (한국가스학회)
권호 표지
DOI QR코드

DOI QR Code

Deep Learning(CNN) based Worker Detection on Infrared Radiation Image Analysis

딥러닝(CNN)기반 저해상도 IR이미지 분석을 통한 작업자 인식

  • Received : 2018.10.18
  • Accepted : 2018.11.24
  • Published : 2018.12.31
Download PDF
⟨ Previous Next ⟩

Abstract

worker-centered safety management for hazardous areas in the plant is required. The causes of gas accidents in the past five years are closely related to the behavior of the operator, such as careless handling of the user, careless handling of the suppliers, and intentional, as well as equipment failure and accident of thought. In order to prevent such accidents, real-time monitoring of hazardous areas in the plant is required. However, when installing a camera in a work space for real-time monitoring, problems such as human rights abuse occur. In order to prevent this, an infrared camera with low resolution with low exposure of the operator is used. In real-time monitoring, image analysis is performed using CNN algorithm, not human, to prevent human rights violation.

플랜트 내 위험지역의 안전을 위해 작업자 중심의 안전관리가 필요하다. 최근 5년간 가스 사고의 원인은 시설 노후 및 장비고장 뿐만 아니라, 사용자의 취급부주의나 고의사고, 공급자 취급부주의 등 작업자의 행동에 밀접한 관련이 있다. 이와 같은 사고를 미연에 방지하기 위해서, 플랜트 내 위험지역에 대한 실시간 모니터링이 필요로 하다. 하지만 실시간 모니터링을 위해서 작업(근로)공간에 카메라 설치 시, 인권침해와 같은 문제가 발생한다. 이를 방지하기 위해서 작업자의 신원 노출이 적은 저해상도의 Infrared 카메라를 이용한다. 또한 실시간 모니터링 시, 사람이 아닌 CNN알고리즘을 이용하여 이미지 분석을 통하여 인권침해 문제를 예방한다.

Keywords

GSGSBE_2018_v22n6_8_f0001.png 이미지

Fig 1. Structure of AlexNet, ImageNet Classification with Deep Convolutional Neural Networks[9]

GSGSBE_2018_v22n6_8_f0002.png 이미지

Fig. 2. Risk calculation process

GSGSBE_2018_v22n6_8_f0003.png 이미지

Fig. 3. Fault Tree Analysis

GSGSBE_2018_v22n6_8_f0004.png 이미지

Fig. 4. Example of a hazard indication in the workplace 1

GSGSBE_2018_v22n6_8_f0005.png 이미지

Fig. 5. Example of a hazard indication in the workplace 2

GSGSBE_2018_v22n6_8_f0006.png 이미지

Fig. 6. IR image analysis process

GSGSBE_2018_v22n6_8_f0007.png 이미지

Fig. 7. data set composition

GSGSBE_2018_v22n6_8_f0008.png 이미지

Fig. 8. Preprocessing image not required

GSGSBE_2018_v22n6_8_f0009.png 이미지

Fig. 9. Preprocessing

GSGSBE_2018_v22n6_8_f0010.png 이미지

Fig. 10. Image requiring preprocessing

GSGSBE_2018_v22n6_8_f0011.png 이미지

Fig. 11. First processing of images

GSGSBE_2018_v22n6_8_f0012.png 이미지

Fig. 12. Image removed from background

GSGSBE_2018_v22n6_8_f0013.png 이미지

Fig. 13. Location detection of object in image

GSGSBE_2018_v22n6_8_f0014.png 이미지

Fig. 14. Object classification

GSGSBE_2018_v22n6_8_f0015.png 이미지

Fig. 15. AlexNet model reconfigured

GSGSBE_2018_v22n6_8_f0016.png 이미지

Fig. 16. Accuracy and loss graphs

GSGSBE_2018_v22n6_8_f0017.png 이미지

Fig. 17. As a result of Bounding Box, The red box area is the worker, the green box area is the gas cylinder.

Table 1. Gas demand per year for the last five years as of 2017, 2017 Gas Statistics of Korea Gas Safety Corporation[1]

GSGSBE_2018_v22n6_8_t0001.png 이미지

Table 2. Cause of accident by year for last 5 years as of 2017, 2017 Gas Year Book of Korea Gas Safety Corporation[2]

GSGSBE_2018_v22n6_8_t0002.png 이미지

References

  1. Korea Gas Safety Cooperation,"Gas accident statics for 2017", Korea Gas Safety Cooperation, (2018)
  2. Korea Gas Safety Cooperation, "gasyearbook 2017", Korea Gas Safety Cooperation, (2018)
  3. Oh, J. S.,Sung, G. J., Kim, Y. D."Developing Network Infrastructure and Smart Service for Safety Management of City-gas Facilities,", KiGAS, Vol. 15(1), 46-53, (2011)
  4. Donald, C."How many monitors should a CCTV operator view.", CCVT Image, 355, (2005)
  5. Kang, S. H, Kim H. S, "Analysis of Privacy Issues by the Diffusion of Video Information Processing System.", Internet & Security Focus 4, 45-65, (2014).
  6. National Human Right Commission of Korea, "Human Rights Commission of Korea, Korea, has launched an analysis on trends of grievance and counseling regarding CCTV.", press release, (2011)
  7. National Human Right Commission of Korea, "Use of supervision of working attitudes other than CCTVs violates human rights.", press release, (2017)
  8. Moon, S. E, et al."Trends in Machine Learning and Deep Learning Technology.", The Korean Institute of Communication Sciences, 33(10), 49-55, (2016)
  9. Krizhevsky, A., Sutskever, I., Hinton, Geoffrey E. "Imagenet classification with deep convolutional neural networks.", In:Advances in neural information processing systems, 1097-1105, (2012)
  10. Szegedy, Christian, et al. "Going deeper with convolutions." Proceedings of the IEEE conference on computer vision and pattern recognition, 1-9, (2015)
  11. He, K. et al. "Deep residual learning for image recognition." Proceedings of the IEEE conference on computer vision and pattern recognition. 770-778, (2016)
  12. Simonyan, K., Zisserman, K. "Very deep convolutional networks for large-scale image recognition." arXiv preprint arXiv:1409.1556, (2014)
  13. Lee, S. J, Jung D. J, Ahn J. H. "Memory Bandwidth Analysis of Various Convolutional Layer Types on Convolutional Neural Networks", KIISE, 1409-1411, (2017)
  14. Yosinski, J, et al. "Understanding neural networks through deep visualization." arXiv preprint arXiv:1506.06579, (2015)
  15. Girshick, R. "Fast r-cnn." Proceedings of the IEEE international conference on computer vision, 1440-1448, (2015)
  16. Dai, J., Li, Y., He, K. "R-fcn: Object detection via region-based fully convolutional networks." Advances in neural information processing systems. 379-387. (2016)
  17. Redmon, J., Divvala, S., Girshick, R., Farhadi, A. "You only look once: Unified, real-time object detection." Proceedings of the IEEE conference on computer vision and pattern recognition. 779-788. (2016)
  18. Liu, W., Anguelov, D., Szegedy, C., Reed, S., Fu, C. Y., Berg, A. C."Ssd: Single shot multibox detector."conference on computer vision. Springer, Cham, 21-37, (2016)

Cited by

  1. 도시가스 누출에 따른 스마트 홈 안전 플랫폼 프로그램 구축 vol.23, pp.6, 2018, https://doi.org/10.7842/kigas.2019.23.6.97
  2. 딥러닝을 활용한 전략물자 판정 지원도구 개발에 대한 연구 vol.30, pp.6, 2020, https://doi.org/10.13089/jkiisc.2020.30.6.967