Модификация архитектуры RTMPose с механизмом внимания CBAM для распознавания позы операторов промышленных площадок при окклюзиях средствами индивидуальной защиты

S A Filichkin; S V Vologdin

doi:10.22213/2410-9304-2026-3-83-88

Authors

S. A. Filichkin Kalashnikov Izhevsk State Technical University
S. V. Vologdin Kalashnikov Izhevsk State Technical University

DOI:

https://doi.org/10.22213/2410-9304-2026-3-83-88

Keywords:

pose, RTMPose, CBAM, attention, occlusions, ppe, safety, surveillance

Abstract

Industrial video monitoring must robustly recognize a worker's pose even when part of the skeleton keypoints is occluded by personal protective equipment (PPE) - a helmet, vest, or gloves - whereas baseline pose-estimation networks lose accuracy noticeably on such frames reducing the reliability of automatic hazard detection. The relevance of the task is determined by the high level of occupational injuries and the need for early detection of safety violations. This paper proposes a modification of the RTMPose architecture, named OA-RTMPose (occlusion-aware): the PoseSimCC backbone is augmented with a CBAM channel-spatial attention module, the loss function is extended with a per-keypoint visibility prediction, and training uses synthetic occlusion augmentation imitating typical PPE items. The joint action of channel-spatial attention, a penalty for invisible keypoints, and targeted augmentation makes the network focus on visible skeleton parts instead of hallucinating hidden ones. The experimental base comprised a subset of the COCO 2017 dataset, an additional set of industrial frames, and the Le2i Fall Detection Dataset. A substantial improvement in the robustness of pose estimation to occlusion was achieved: on the subset of occluded cases, the AP75 metric increased by 72.7% relative to the baseline model, while an analytical projection onto typical industrial occlusions yields an expected advantage of about +17.1%; while the model preserves real-time video processing. The obtained results indicate the feasibility of applying the proposed modification in intelligent industrial-safety monitoring systems, where occlusion-robust worker-pose recognition increases the reliability of early hazard detection and reduces the false-alarm rate.

Author Biographies

S. A. Filichkin, Kalashnikov Izhevsk State Technical University

Post-graduate

S. V. Vologdin, Kalashnikov Izhevsk State Technical University

DSc in Engineering, Associate Professor, Information Systems Department

References

Loktev D. A., Illarionova L. A., Loktev A. A. Automation of the behavioral safety audit of a transport operator based on the recognition of his emotional state in real time // 2023 5th International Youth Conference on Radio Electronics, Electrical and Power Engineering (REEPE). 2023. P. 1-5. DOI: 10.1109/REEPE57272.2023.10086820.

López L., Suárez-Ramírez J., Alemán-Flores M., Monzón N. Automated PPE compliance monitoring in industrial environments using deep learning-based detection and pose estimation // Automation in Construction. 2025. Vol. 176. Art. 106231. DOI: 10.1016/j.autcon.2025.106231.

Sun K., Xiao B., Liu D., Wang J. Deep high-resolution representation learning for human pose estimation // Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2019. P. 5686-5696. DOI: 10.1109/CVPR.2019.00584.

Ковалев В. В., Сергеев Н. Е. Реализация сверточных нейронных сетей на встраиваемых устройствах с ограниченным вычислительным ресурсом // Известия Южного федерального университета. Технические науки. 2021. № 6. С. 64-72. DOI: 10.18522/2311-3103-2021-6-64-72.

Sandler M., Howard A., Zhu M., Zhmoginov A., Chen L.-C. MobileNetV2: inverted residuals and linear bottlenecks // Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2018. P. 4510-4520. DOI: 10.1109/CVPR.2018.00474.

Jiang T., Lu P., Zhang L., Ma N., Han R., Lyu C., Li Y., Chen K. RTMPose: real-time multi-person pose estimation based on MMPose. 2023. URL: https://arxiv.org/abs/2303.07399 (дата обращения: 18.05.2026). DOI: 10.48550/arXiv.2303.07399.

Li Y., Yang S., Liu P., Zhang S., Wang Y., Wang Z., Yang W., Xia S.-T. SimCC: a simple coordinate classification perspective for human pose estimation // Proc. European Conference on Computer Vision (ECCV). 2022. P. 89-106. DOI: 10.1007/978-3-031-20068-7_6.

Метод трёхмерного захвата движения человека на основе компьютерного зрения / А. Д. Обухов, Д. Л. Дедов, Е. О. Суркова, И. Л. Коробова // Advanced Engineering Research (Rostov-on-Don). 2023. Т. 23, № 3. С. 317-328. DOI: 10.23947/2687-1653-2023-23-3-317-328.

Lin T.-Y., Maire M., Belongie S., Hays J., Perona P., Ramanan D., Dollár P., Zitnick C. L. Microsoft COCO: common objects in context // Proc. European Conference on Computer Vision (ECCV). 2014. P. 740-755. DOI: 10.1007/978-3-319-10602-1_48.

Andriluka M., Pishchulin L., Gehler P., Schiele B. 2D human pose estimation: new benchmark and state-of-the-art analysis // Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2014. P. 3686-3693. DOI: 10.1109/CVPR.2014.471.

Sárándi I., Linder T., Arras K. O., Leibe B. MeTRAbs: metric-scale truncation-robust heatmaps for absolute 3D human pose estimation // IEEE Transactions on Biometrics, Behavior, and Identity Science. 2021. Vol. 3, No. 1. P. 16-30. DOI: 10.1109/TBIOM.2020.3037257.

Cheng Y., Yang B., Wang B., Yan W., Tan R. T. Occlusion-aware networks for 3D human pose estimation in video // Proc. IEEE/CVF International Conference on Computer Vision (ICCV). 2019. P. 723-732. DOI: 10.1109/ICCV.2019.00081.

Liu W., Bao Q., Sun Y., Mei T. Recent advances of monocular 2D and 3D human pose estimation: a deep learning perspective // ACM Computing Surveys. 2022. Vol. 55, No. 4. Art. 80. DOI: 10.1145/3524497.

Hu J., Shen L., Sun G. Squeeze-and-Excitation Networks // Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2018. P. 7132-7141. DOI: 10.1109/CVPR.2018.00745.

Woo S., Park J., Lee J.-Y., Kweon I. S. CBAM: Convolutional Block Attention Module // Proc. European Conference on Computer Vision (ECCV). 2018. P. 3-19. DOI: 10.1007/978-3-030-01234-2_1.

Wang Q., Wu B., Zhu P., Li P., Zuo W., Hu Q. ECA-Net: efficient channel attention for deep convolutional neural networks // Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2020. P. 11531-11539. DOI: 10.1109/CVPR42600.2020.01155.

Hou Q., Zhou D., Feng J. Coordinate attention for efficient mobile network design // Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2021. P. 13708-13717. DOI: 10.1109/CVPR46437.2021.01350.

Sun X., Xiao B., Wei F., Liang S., Wei Y.Integral human pose regression // Proc. European Conference on Computer Vision (ECCV). 2018. P. 536-553. DOI: 10.1007/978-3-030-01231-1_33.

Charfi I., Miteran J., Dubois J., Atri M., Tourki R. Definition and performance evaluation of a robust SVM-based fall detection solution // Proc. 8th International Conference on Signal Image Technology and Internet Based Systems (SITIS). 2012. P. 218-224. DOI: 10.1109/SITIS.2012.155.

Филичкин С. А., Вологдин С. В. Разработка системного подхода, новых методов и алгоритмов для распознавания различных поведенческих функций человека // Вестник Воронежского института ФСИН России. 2026. № 1. С. 149-158.