Demiryolu Ortamında Nesne Tespiti için Derin Öğrenme Yöntemlerinin Geliştirilmesi ve Zed Kamerası ile Mesafe Ölçümü

Yıl 2025, Sayı: 22, 10 - 24, 31.07.2025

Muhammed Amir Elmuhammedcebben , İlhan Aydın , Mehmet Sevi

https://doi.org/10.47072/demiryolu.1645019

Öz

Demiryolu güvenliğinin sağlanması, dinamik ortamlarda doğru ve gerçek zamanlı nesne tespiti gerektiren kritik bir mühendislik problemidir. Bu çalışmada, demiryolu izleme süreçlerini iyileştirmek amacıyla derin öğrenme tabanlı nesne tespiti modelleri geliştirilmiş ve RailSem19 veri seti kullanılarak YOLOv8, YOLOv9, YOLOv11 ve Faster R-CNN modelleri eğitilerek entegre edilmiştir. Ayrıca, nesnelerin mesafe ölçümlerinin hassasiyetini artırmak için ZED stereo kamerası kullanılmıştır. Deneysel bulgular, YOLOv9 modelinin doğruluk ve işlem verimliliği açısından en iyi dengeyi sağladığını ortaya koymakta olup, bu modelin ortalama doğruluk (mAP50) değeri %52,14 olarak hesaplanmış ve çıkarım süresi optimize edilmiştir. Önerilen yöntem, otonom demiryolu sistemleri ve güvenlik izleme uygulamaları için önemli bir potansiyel taşımaktadır. Gelecekteki çalışmalar, veri artırma teknikleri, yapay zekâ destekli yorumlanabilirlik ve çoklu sensör füzyonu ile modelin dayanıklılığını artırmaya odaklanacaktır.

Anahtar Kelimeler

Derin öğrenme , Demiryolu güvenliği , Nesne algılama , Mesafe ölçümü , RailSem19 , ZED kamera

Developing Deep Learning Methods for Object Detection and Using Zed Camera for Distance Measurement in Railway Environment

Öz

Ensuring railway safety is a critical engineering problem that requires accurate and real-time object detection in dynamic environments. In this study, deep learning-based object detection models are developed to improve railway monitoring processes, and YOLOv8, YOLOv9, YOLOv11, and Faster R-CNN models are trained and integrated using RailSem19 dataset. In addition, a ZED stereo camera is used to increase the precision of distance measurements of objects. Experimental findings reveal that the YOLOv9 model provides the best balance in terms of accuracy and computational efficiency, the average accuracy (mAP50) value of this model is calculated as 52.14% and the inference time is optimized. The proposed method has significant potential for autonomous railway systems and safety monitoring applications. Future studies will focus on improving the robustness of the model with data augmentation techniques, artificial intelligence-supported interpretability, and multi-sensor fusion.

Anahtar Kelimeler

Deep learning , Railway safety , Object detection , Distance measurement , RailSem19 , ZED camera

Kaynakça

• [1] Z. Zhang, P. Chen, Y. Huang, L. Dai, F. Xu, and H. Hu, “Railway obstacle intrusion warning mechanism integrating YOLO-based detection and risk assessment,” J Ind Inf Integr, vol. 38, pp. 100571, Mar. 2024, doi: 10.1016/J.JII.2024.100571
• [2] A. Katham Mtashre, D. Mohsin Kareem, and Z. Abd Al-Abbas Muhsen, “Enhancing Object Detection Techniques Through Transfer Learning and Pre-trained Models,” Journal of Engineering Sciences and Information Technology (JESIT), vol. 3, no. 8, pp. 39–45, Sep, 2024, doi: 10.26389/AJSRP.K270724
• [3] A. Sychugov, V. Miheychikov, and M. Chernyshov, “Application of Neural Networks for Object Recognition in Railway Transportation,” Izvestiâ Peterburgskogo universiteta putej soobŝeniâ, vol. 20, no. 2, pp. 478–491, Jun. 2023, doi: 10.20295/1815-588X-2023-2-478-491
• [4] J. Zhao and M. Su, “Comparative Performance Analysis of Single-Shot Detector and Faster R-CNN for Object Detection,” Computer Science, Engineering and Information Technology, pp. 209–222, Oct. 2024, doi: 10.5121/CSIT.2024.141919
• [5] M. Sevi and İ. Aydın, “Detection of Foreign Objects Around the Railway Line with YOLOv8,” Computer Science, no. IDAP-2023, pp. 19–23, Oct. 2023, doi: 10.53070/BBD.1346317
• [6] C. Yong, W. Zhen, and Z. Fangchun, “Railway foreign object tracking and detection with spatial positioning and feature generalization enhancement,” Journal of Applied Artificial Intelligence, vol. 1, no. 3, pp. 260–274, Oct. 2024, doi: 10.59782/AAI.V1I3.329
• [7] N. Bilous, V. Malko, M. Frohme, and A. Nechyporenko, “Comparison of CNN-Based Architectures for Detection of Different Object Classes,” AI 2024, vol. 5, no. 4, pp. 2300–2320, Nov. 2024, doi: 10.3390/AI5040113
• [8] S. Ning, F. Ding, and B. Chen, “Research on the Method of Foreign Object Detection for Railway Tracks Based on Deep Learning,” Sensors, vol. 24, no. 14, p. 4483, Jul. 2024, doi: 10.3390/S24144483
• [9] A. Selvi, S. G. Ragul Ram, K. Sharanraj, and K. Thirukumaran, “AI Precision on Rails Advanced Object Recognition for Train Track Safety - A Survey,” Proceedings - 2024 5th International Conference on Intelligent Communication Technologies and Virtual Mobile Networks, Tirunelveli, India, 2024, pp. 388–394.
• [10] H. Xianzhi, “Rail traffic distance measurement system and method,” 2025. [Online]. Available: https://typeset.io/papers/rail-traffic-distance-measurement-system-and-method-3wci1a2uub
• [11] M. Nitti et al., “3d stereo reconstruction of train paths for supporting maintenance operations,” EGU General Assembly Conference Abstracts, Mar. 2020, doi: 10.5194/EGUSPHERE-EGU2020-20265
• [12] H. K. Drizi and M. Boukadoum, “CNN Model with Transfer learning and Data Augmentation for Obstacle Detection in Rail Systems,” Proceedings - IEEE International Symposium on Circuits and Systems, Singapore, Singapore, 2024, pp. 1-5.
• [13] X. Wang, J. Yin, F. Pu, X. Chen, A. D’Ariano, and T. Tang, “A GAN-based Deep Learning Model for the Object Detection of Autonomous High-Speed Trains,” Proceedings - 2023 China Automation Congress, Chongqing, China, 2023, pp. 4393–4398.
• [14] M. F. Nicolas and D. B. Megherbi, “Hidden Challenge in Deep-Learning Real-Time Object Detection on Edge Devices,” Midwest Symposium on Circuits and Systems, Springfield, MA, USA, 2024, pp. 547–551.
• [15] H. Song, X. Song, M. Zhou, L. Liu, and H. Dong, “Unsupervised Deep Learning-Based Point Cloud Detection for Railway Foreign Object Intrusion,” 2024 IEEE 19th Conference on Industrial Electronics and Applications, Kristiansand, Norway, 2024, pp. 1-6.
• [16] M. A. R. Alif and M. Hussain, “Lightweight Convolutional Network with Integrated Attention Mechanism for Missing Bolt Detection in Railways,” Metrology, vol. 4, no. 2, pp. 254–278, May 2024. doi: 10.3390/METROLOGY4020016
• [17] N. T. Pham, A. N. Timofeev, and I. S. Nekrylov, “Study of the errors of stereoscopic optical-electronic system for railroad track position,” In Optical Measurement Systems for Industrial Inspection XI, vol. 11056, pp. 622–635, Jun. 2019, doi: 10.1117/12.2526081
• [18] O. Zendel, M. Murschitz, M. Zeilinger, D. Steininger, S. Abbasi, and C. Beleznai, “RailSem19: A dataset for semantic rail scene understanding,” IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops, vol. 2019-June, pp. 1221–1229, Jun. 2019, doi: 10.1109/CVPRW.2019.00161
• [19] N. Jegham, C. Y. Koh, M. Abdelatti, and A. Hendawi, “YOLO Evolution: A Comprehensive Benchmark and Architectural Review of YOLOv12, YOLO11, and Their Previous Versions,” arXiv preprint arXiv:2311.10041, 2023.
• [20] S. Y. Rhyou, M. Yu, and J. C. Yoo, “Mixture of expert-based SoftMax-weighted box fusion for robust lesion detection in ultrasound imaging,” Diagnostics, vol. 15, no. 5, p. 588, 2025.
• [21] Y. Ji, D. Zhang, Y. He, J. Zhao, X. Duan, and T. Zhang, “Improved YOLO11 algorithm for insulator defect detection in power distribution lines,” Electronics, vol. 14, no. 6, p. 1201, 2025.
• [22] F. Saeed, A. Paul, and S. Rho, “Faster R-CNN based fault detection in industrial images,” in Trends in Artificial Intelligence Theory and Applications. Artificial Intelligence Practices: Proc. 33rd Int. Conf. on Industrial, Engineering and Other Applications of Applied Intelligent Systems (IEA/AIE 2020), Kitakyushu, Japan, Sep. 22–25, 2020, pp. 280–287, Springer.
• [23] O. Rodríguez-Abreo, M. A. Quiroz-Juárez, I. Macías-Socarras, J. Rodríguez-Reséndiz, J. M. Camacho-Pérez, G. Carcedo-Rodríguez, and E. Camacho-Pérez, “Automatic detection of railway faults using neural networks: A comparative study of transfer learning models and YOLOv11,” Infrastructures, vol. 10, no. 1, p. 3, 2024.
• [24] S. Y. Nikouei, Y. Chen, S. Song, R. Xu, B. Y. Choi, and T. Faughnan, “Smart surveillance as an edge network service: From Haar-cascade, SVM to a lightweight CNN,” in Proc. 2018 IEEE 4th Int. Conf. on Collaboration and Internet Computing (CIC), Philadelphia, PA, USA, Oct. 2018, pp. 256–265.