TMK Research Center, Moscow, Russia
I. Yu. Pyshmintsev, Dr. Eng., Prof., General Director
E. A. Shkuratov, Cand. Eng., Head of the Digitalization and Artificial Intelligence Technologies Department, e-mail: evgeniy.shkuratov@tmk-group.com
I. V. Kosmin, Research Engineer, Digitalization and Artificial Intelligence Technologies Department
S. K. Rosolenko, Engineer, Digitalization and Artificial Intelligence Technologies Department
G. A. Yashin, Development Engineer, Digitalization and Artificial Intelligence Technologies Department

This paper presents the development of an adaptive system for monitoring the geometric parameters of hot-rolled pipes based on machine vision technologies. A software architecture is proposed that automatically retunes image-processing algorithms to changing rolling regimes and observation conditions. A systematic analysis is carried out of the factors affecting measurement accuracy at high temperatures, in particular temperature gradients, glare, scale formation, equipment vibrations, and variations in billet transport speed. An applied classification of imaging modes and thermal-impact levels is developed for the key stages of a tube rolling mill line, enabling a rational choice between static and dynamic measurement methods and the configuration of data-processing trajectories. The algorithmic core combines sparse and dense optical-flow estimators for frame-to-frame displacement summation, neural end-face localization with detection of entry/exit moments into the control zone, and adaptive region-of-interest (ROI) formation with glare/scale segmentation. In addition, quantile filtering and forward–backward consistency checks are employed to reduce outliers and systematic calibration errors. The mean error of inter-frame shift estimation did not exceed 0.1 pixels, ensuring high accuracy in computing geometric parameters. Industrial deployment of the system demonstrated improved accuracy and repeatability of real-time length and diameter measurement. The microservice architecture provided system fault tolerance, scalability, and seamless integration with existing digital infrastructures of industrial enterprises, thereby creating a foundation for expanding technological control functions and introducing adaptive process control across the entire tube rolling mill line.

1. Shkuratov E. A., Ulman N. V., Levshin D. V., Pyankov K. P. Basic principles and approaches to creating a predictive model to minimize equipment downtime in the production of pipes on a continuous mill. Metallurg. 2025. No. 5. pp. 100–105.
2. Shkuratov E. A., Belov V. D., Kodos N. N., Kosmin I. V., Guseynov E. R. et al. Program for determining the positions of rolls of a continuous mill in the production of pipes Roll Position. State registration No. 2024690777. Applied: 04.12.2024. Published: 17.12.2024.
3. Pyshmintsev I. Yu., Shkuratov E. A., Yashin G. A., Kosmin I. V. et al. Development of a hardware and software complex for in-line pipe inspection in continuous production. Metallurg. 2025. No. 6. pp. 82–87.
4. Kosmin I. V., Shkuratov E. A., Bushin R. O., Pyankov K. P., Panasenko O. A. Increasing the design accuracy of pipe-rolling equipment as a result of using 3D laser systems and mathematical models for processing. Metallurg. 2023. No. 7. pp. 97–101.
5. Shkuratov E. A., Vydrin A. V. Improvement of the technology for hollow shell longitudinal rolling in continuous rolling mills. Chernye Metally. 2017. No. 3. pp. 42–46.
6. Vydrin A. V. Development of technologies for production of pipes with specified properties based on modern research methods. Pipes-2021: Proceedings of the XXIV International Scientific and Practical Conference, Chelyabinsk, 20–22.09.2021. Part 1. Chelyabinsk : RusNITI, 2021. pp. 117–121.
7. Song X., Cao S., Zhang J., Hou Z. Steel surface defect detection algorithm based on YOLOv8. Electronics. 2024. Vol. 13. 988. DOI: 10.3390/electronics13050988
8. Yang W., Zheng J. Real-time face detection based on YOLO. Proc. of ICKII. 2018. pp. 221–224. DOI: 10.1109/ICKII.2018.8569109
9. Aboyomi D. D., Daniel C. A comparative analysis of modern object detection algorithms: YOLO vs. SSD vs. Faster R-CNN. ITEJ. 2023. Vol. 8, No. 2. pp. 96–106. DOI: 10.24235/itej.v8i2.123
10. Nayar S. K. et al. Real-time 100 object recognition system. Proc. IEEE Int. Conf. on Robotics and Automation. 1996. Vol. 3. pp. 2321–2325.
11. Maturana D., Scherer S. A. VoxNet: A 3D Convolutional neural network for real-time object recognition. Proc. IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS). 2015. pp. 922–928.
12. Lucas B. D., Kanade T. An iterative image registration technique with an application to stereo vision. Proc. of Imaging Understanding Workshop. 1981. pp. 121–130.
13. Baker S., Matthews I. Lucas–Kanade 20 years on: A unifying framework. International Journal of Computer Vision. 2004. Vol. 56, No. 3. pp. 221–255.
14. Bouguet J.-Y. Pyramidal implementation of the Lucas–Kanade feature tracker: description of the algorithm. Intel Corporation, Microprocessor Research Labs, 2000. 9 p.
15. Farnebäck G. Two-frame Motion Estimation Based on Polynomial Expansion. Proc. 13^th Scandinavian Conference on Image Analysis (SCIA). 2003. pp. 363–370.
16. Bruhn A., Weickert J., Schnörr C. Lucas/Kanade meets Horn/Schunck: Combining local and global optic flow methods. International Journal of Computer Vision. 2005. Vol. 61. pp. 211–231.