Hilmi Saygın SUCUOĞLU, İsmail BÖĞREKCİ, Pınar DEMİRCİOĞLU

FASTER R-CNN MODELİ KULLANARAK GERÇEK ZAMANLI YANGIN ALGILAMA

Bu çalışmanın amacı, Faster R-CNN kullanarak gerçek zamanlı bir yangın dedektörü geliştirmektir. Sistemin makine öğrenme süreci için; 1.000 resim (yangın ve yangın sahneleri dahil), eğitim ve doğrulama için sırasıyla %80 ve 20 oranında kullanılmıştır. Makine öğrenme işlemi, NVidia GeForce GTX 1070 Ti'nin 17 GB dahili belleğe sahip özelliklerine sahip bir sistem kullanılarak gerçekleştirildi. Anaconda sanal ortamına gerekli yazılımlar (Python 3.5, Tensorflow 1.13.1, OpenCV, CUDA-cuDNN araç setleri) kuruldu. Görüntülerdeki yangın sahneleri, LabelImg yazılımı kullanılarak yangın ve yangın olmayan olarak etiketlendi. Eğitim sürecinin ölçümleri Tensorboard'dan alınmıştır. Eğitimde 40.000'lik adımlarla toplam zarar değeri 2'den 0,02'ye düşürüldü. Lost fonksiyonu 0.05 seviyesinden düşük olduğu için, çıkarım grafiği donmuş ve yangın kaynağını tespit etmek için dışa aktarılmıştır. Geliştirilen gerçek zamanlı yangın dedektörü modeli, yangın kaynağı olarak çakmak kullanılarak gerçek zamanlı olarak test edildi. Test sonuçlarında; doğruluğun % 99'u gelişmiş Faster R-CNN yangın dedektörü modeli kullanılarak elde edildi.

Anahtar Kelimeler:

Faster R-CNN, Loss Function, Real time Fire Detection, Tensorflow and Tensorboard

REAL TIME FIRE DETECTION USING FASTER R-CNN MODEL

The aim of this study is to develop a real time fire detector using Faster R-CNN (Faster region-based convolutional neural network). For machine learning process of the system; 1,000 images (including fire and non-fire scenes) 80 and 20% for training and validation, respectively were used. The machine learning process was conducted using a system with the specifications of NVidia GeForce GTX 1070 Ti with 17 GB onboard memory. The required environments and tools (Python 3.5, Tensorflow 1.13.1, OpenCV, CUDA-cuDNN toolkits) were installed on the Anaconda virtual environment. The fire scenes on the images were labeled as fire and non-fire using LabelImg software. The metrics of the training process were obtained from the Tensorboard. The total loss value decreased from 2 to 0.02 with the steps of 40,000 at training. As the loss function was lower than the level of 0.05, inference graph was frozen and exported to detect the fire source. The developed real time fire detector model was tested in real time using lighter as fire source. In the test results; the 99% of accuracy was obtained using developed Faster R-CNN fire detector model.

Keywords:

Faster R-CNN, Loss Function, Real time Fire Detection, Tensorflow and Tensorboard,

PDF

___

Referans 1. Tan, C.F., Liew, S.M., Alkahari, M.R., Ranjit, S.S.S., Said, M.R., Chen W., Rauterberg, G.W.M., Sivarao D.S. “Fire Fighting Mobile Robot: State of the Art and Recent Development”, Australian Journal of Basic and Applied Sciences, Vol.10, Pages 220-230, 2013.
Referans 2. Luo, R.C., Su, K.L. “Autonomous Fire-Detection System Using Adaptive Sensory Fusion for Intelligent Security Robot”, IEEE/ASME Transactions on Mechatronics, Vol. 12, Pages 274-281, 2007.
Referans 3. Khoon, T.N., Sebastian, P., Saman, A.B.S. “Autonomous Fire Fighting Mobile Platform”, Procedia Engineering, Vol. 41, Pages 1145-1153, 2012.
Referans 4. Chang, P.H., Kang, Y.H., Cho, G.R., Kim, J.H., Jin, M., Lee, J. “Control Architecture Design for a Fire Searching Robot using Task Oriented Design Methodology”, SICE-ICASE International Joint Conference, IEEE, Pages 3126 - 3131, Busan, 2006.
Referans 5. Kim, Y. D., Kim, Y. G., Lee, S. H., Kang, J. H., An, J. “Portable Fire Evacuation Guide Robot System”, The 2009 IEEE/RSJ International Conference on Intelligent Robots and Systems October 11-15, St. Louis, USA, 2009.
Referans 6. Roberto, G.F., Branco, K.C., Machado, J.M., Pinto, A.R. “Local Data Fusion Algorithm for Fire Detection through Mobile Robot”, Test Workshop (LATW), 14th Latin American, 1-6, Cordoba, 2013.
Referans 7. Kumar, P.S, Ratheesh, B.R., Gobinath, B., Kumaran, K.M., Venkataraman, S. “Gesture Controlled Fire Extinguisher Robot with Audio and Video Capture”, IOSR Journal of Electronics and Communication Engineering, Pages 101-105, 2007.
Referans 8. Necsulescu, D. S., ur Rehman, A., & Sasiadek, J. “Fire detection robot navigation using modified voting logic”, In Informatics in Control Automation and Robotics (ICINCO), 2014 11th International Conference on IEEE, No. 1, Pages 140-146, 2014.
Referans 9. Frizzi, S., Kaabi, R., Bouchouicha, M., Ginoux, J. M., Moreau, E., Fnaiech, F. “Convolutional Neural Network for Video Fire and Smoke Detection”, IECON 2016 - 42nd Annual Conference of the IEEE Industrial Electronics Society, Pages 877-882, 2016.
Referans 10. He, K., Zhang, X., Ren, S., Sun, J. “Deep residual learning for image recognition”, In Proceedings of the IEEE conference on computer vision and pattern recognition, Pages 770-778, 2016.
Referans 11. Kim, Y. J., Kim, E. G. “A Study on Fire Detection Using Faster R-CNN and ResNet”, Information, Vol.21, Issue 1, Pages173-180, 2018.
Referans 12. Celik, T., Demirel, H. “Fire detection in video sequences using a generic color model”, Fire Safety Journal, Vol. 44, Pages 147-158, 2009.
Referans 13. Horng, W. B., Peng, J. W., Chen, C. Y. “A new image-based real-time flame detection method using color analysis”, In Networking, Sensing and Control, Proceedings, 2005 IEEE, Pages 100-105, 2005.
Referans 14. Töreyin, B. U., Dedeoglu, Y., Güdükbay, U., Cetin, A. E. “Computer vision based method for real-time fire and flame detection”, Pattern recognition letter, Vol.27, Pages 49-58, 2006.
Referans 15. Kim, Y. J., Kim, E. G. “Image based fire detection using convolutional neural network”, Journal of the Korea Institute of Information and Communication Engineering, Vol. 20, Pages 1649-1656, 2016.
Referans 16. Understanding of Convolutional Neural Network (CNN) — Deep Learning, https://medium.com/@RaghavPrabhu/understanding-of-convolutional-neural-network-cnn-deep-learning-99760835f148, October 21, 2019.
Referans 17. What Is a Convolutional Neural Network? https://ch.mathworks.com/solutions/deep-learning/convolutional-neural-network.html, October 21, 2019.
Referans 18. Faster RCNN Object detection, https://towardsdatascience.com/faster-rcnn-object-detection-f865e5ed7fc4, October 21, 2019.