Modern Transportation Systems and TechnologiesModern Transportation Systems and TechnologiesИнновационные транспортные системы и технологии2782-3733Eco-Vector70715310.17816/transsyst707153Original studiesОригинальные статьиResearch ArticleNeural network integration of transport demand and transport supply features for highway traffic load assessmentМетодика нейросетевой интеграции транспортного спроса и признаков транспортного предложения для оценки загрузки автомобильных дорогhttps://orcid.org/0000-0002-6855-06884337-7527KotovAndrey A.КотовАндрей Александрович
Russian Federation

Associate Professor, CEO

доцент МАДИ, генеральный директор

gt@madi.ruhttps://orcid.org/0009-0002-5700-07336379-1134KozhevnikovSergey N.КожевниковСергей Николаевич
Russian Federation

GIS / Data science analyst, President

инженер-исследователь, президент

skozhevnikov@cosmogeology.comthe Moscow Automobile and Road Construction State Technical UniversityМосковский автомобильно-дорожный государственный технический университетGEOTRANS LLCООО «ГЕОТРАНС»COSMOGEOLOGY LLCООО «КОСМОГЕОЛОГИЯ»060720261222682970405202623052026Copyright ©; 2026, Kotov A.A., Kozhevnikov S.N.Copyright ©; 2026, Котов А.А., Кожевников С.Н.2026Kotov A.A., Kozhevnikov S.N.Котов А.А., Кожевников С.Н.https://creativecommons.org/licenses/by/4.0https://transsyst.ru/transj/article/view/707153

Background: Digital transformation of transport infrastructure requires coordinated accounting of transport demand, transport supply features, and the reliability of source observations. Separate handling of video data, coordinates, road objects, and calculated indicators reduces the reproducibility of highway traffic load assessment.

Aim: This study aims to develop a method for integrating neural network video monitoring, linear referencing, GeoAI processing of spatial data, and a road graph to calculate traffic load together with a reliability index.

Methods: The method combines video recording, GNSS/INS referencing, LiDAR and GIS data, YOLOv12–YOLOv13 and RT-DETRv4 detection, ByteTrack/BoT-SORT tracking, SAM 2–based verification segmentation, and calculation of traffic load estimates that account for observation reliability and confidence intervals.

Results: A 52-hour control video sample produced mAP@50 up to 0.91 for passenger cars, MOTA of 0.78, IDF1 of 0.84, and a mean daytime counting error of 6.4%.

Conclusion: The method provides traceable integration of transport demand, transport supply features, and calculated indicators within a unified spatio-temporal road graph, which serves as a single source of truth for traffic load calculation and verification of disputed fragments.

Обоснование. Управление транспортной инфраструктурой в условиях цифровой трансформации требует согласованного учета транспортного спроса, признаков транспортного предложения и достоверности исходных наблюдений. Раздельное ведение видеоданных, координат, дорожных объектов и расчетных показателей снижает воспроизводимость оценки загрузки автомобильных дорог.

Цель. Разработать методику интеграции нейросетевого видеомониторинга, линейной привязки, GeoAI-обработки пространственных данных и дорожного графа для расчета уровня загрузки с индексом достоверности.

Материалы и методы. Методика объединяет видеосъемку, ГНСС/ИНС-привязку, лидарные и ГИС-данные, детекцию YOLOv12–YOLOv13 и RT-DETRv4, трекинг ByteTrack/BoT-SORT, проверочную сегментацию SAM 2, расчетную оценку загрузки дороги с учетом достоверности наблюдений и доверительного интервала.

Результаты. На контрольной выборке 52 ч видео получены mAP@50 до 0,91 для легковых транспортных средств, MOTA 0,78, IDF1 0,84; средняя ошибка подсчета в дневных условиях составила 6,4%.

Заключение. Методика обеспечивает трассируемое соединение транспортного спроса, признаков транспортного предложения и расчетных показателей в едином пространственно-временном дорожном графе, который используется как единая версия сведений для расчета загрузки и проверки спорных фрагментов.

transport infrastructuretraffic loadcomputer visionYOLOv13RT-DETRv4SAM 2GeoAIroad graphGIS integrationreliability indexтранспортная инфраструктурауровень загрузкикомпьютерное зрениеYOLOv13RT-DETRv4SAM 2GeoAIдорожный графГИС-интеграцияиндекс достоверностиThe article was prepared as part of the state assignment of the Ministry of Science and Higher Education of Russia (topic No. FSFM-2025-0004 Using public neural networks for automated generation of elements of street and road transport infrastructure).Статья подготовлена в рамках государственного задания Минобрнауки России (тема № FSFM-2025-0004 Применение публичных нейросетей для автоматизированного генерирования элементов уличной и дорожно-транспортной инфраструктуры).Federal Highway Administration. Traffic Monitoring Guide. FHWA-PL-17-003. Washington, DC: FHWA; 2016. Accessed: 15.06.26. Available from: https://www.fhwa.dot.gov/policyinformation/tmguide/tmg_fhwa_pl_17_003.pdfFederal Highway Administration. Traffic Monitoring Guide. FHWA-PL-17-003. Washington, DC: FHWA, 2016. Дата обращения: 15.06.26. Режим доступа: https://www.fhwa.dot.gov/policyinformation/tmguide/tmg_fhwa_pl_17_003.pdfAI City Challenge. Official benchmark and datasets for traffic video analytics. 2025. Accessed: 15.06.26. Available from: https://www.aicitychallenge.orgAI City Challenge. Official benchmark and datasets for traffic video analytics, 2025. Дата обращения: 15.06.26. Режим доступа: https://www.aicitychallenge.orgISO/TS 14812:2025. Intelligent transport systems - Vocabulary. ISO; 2025. Accessed: 15.06.26. Available from: https://www.iso.org/standard/85041.htmlISO/TS 14812:2025. Intelligent transport systems - Vocabulary. ISO, 2025. Дата обращения: 15.06.26. Режим доступа: https://www.iso.org/standard/85041.htmlVasil’eva ME, Volkova EM, Romanov AS. Intelligent transport systems in Russian agglomerations: essence, structure and development directions. Innovatsionnye Transportnye Sistemy i Tekhnologii. 2023;9(4):117-128. (In Russ.) doi: 10.17816/transsyst202394117-128 EDN: OLECOSВасильева М.Е., Волкова Е.М., Романов А.С. Интеллектуальные транспортные системы в российских агломерациях: сущность, структура и направления развития // Инновационные транспортные системы и технологии. 2023. Т. 9, № 4. С. 117-128. doi: 10.17816/transsyst202394117-128 EDN: OLECOSAukhadeev AE. Application of artificial intelligence technologies in substantiating algorithms for effective control of the electrical complex of traction equipment of urban electric transport. Innovatsionnye Transportnye Sistemy i Tekhnologii. 2024;10(3):368-389. (In Russ.) doi: 10.17816/transsyst635508 EDN: BDLAKVАухадеев А.Е. Применение технологий искусственного интеллекта при обосновании алгоритмов эффективного управления электротехническим комплексом тягового электрооборудования городского электротранспорта // Инновационные транспортные системы и технологии. 2024. Т. 10, № 3. С. 368-389. doi: 10.17816/transsyst635508 EDN: BDLAKVOstroukh AV, Podberezkin AA, Pronin TsB, et al. Development and training of a domain-specific large language model for automating the extraction and validation of parameters in transport modeling tasks. Sovremennye Naukoemkie Tekhnologii. 2026;(4):317-323. (In Russ.) doi: 10.17513/snt.40767 EDN: NDKGBOОстроух А.В., Подберёзкин А.А., Пронин Ц.Б., и др. Разработка и обучение предметно-ориентированной большой языковой модели для автоматизации извлечения и валидации параметров в задачах транспортного моделирования // Современные наукоемкие технологии. 2026. № 4. С. 317-323. doi: 10.17513/snt.40767 EDN: NDKGBOMatias J. Computer Vision Methods for Vehicle Detection and Tracking. Appl Sci. 2025;15(22):12288. doi: 10.3390/app152212288 EDN: FENTHIMatias J. Computer Vision Methods for Vehicle Detection and Tracking // Applied Sciences. 2025. Vol. 15, No. 22. Article 12288. doi: 10.3390/app152212288 EDN: FENTHIMa C, Xue F. A review of vehicle detection methods based on computer vision. J Intell Connect Veh. 2024;7(1):1-18. doi: 10.26599/JICV.2023.9210019Ma C., Xue F. A review of vehicle detection methods based on computer vision // Journal of Intelligent and Connected Vehicles. 2024. Vol. 7, No. 1. P. 1-18. doi: 10.26599/JICV.2023.9210019Tian Y, Ye Q, Doermann D. YOLOv12: Attention-Centric Real-Time Object Detectors. arXiv preprint. 2025. doi: 10.48550/arXiv.2502.12524Tian Y., Ye Q., Doermann D. YOLOv12: Attention-Centric Real-Time Object Detectors. arXiv preprint, 2025. doi: 10.48550/arXiv.2502.12524Lei M, Li S, Wu Y, et al. YOLOv13: Real-Time Object Detection with Hypergraph-Enhanced Adaptive Visual Perception. arXiv preprint. 2025. doi: 10.48550/arXiv.2506.17733Lei M., Li S., Wu Y., et al. YOLOv13: Real-Time Object Detection with Hypergraph-Enhanced Adaptive Visual Perception. arXiv preprint, 2025. doi: 10.48550/arXiv.2506.17733Ultralytics. YOLO26 Documentation. Ultralytics; 2026. Accessed: 15.06.26. Available from: https://docs.ultralytics.com/models/yolo26Ultralytics. YOLO26 Documentation. Ultralytics, 2026. Дата обращения: 15.06.26. Режим доступа: https://docs.ultralytics.com/models/yolo26Zhao Y., Lv W., Xu S., et al. DETRs Beat YOLOs on Real-time Object Detection // Proceedings of the 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2024. P. 16965-16974. doi: 10.1109/CVPR52733.2024.01605Zhao Y, Lv W, Xu S, et al. DETRs Beat YOLOs on Real-time Object Detection. In: *Proceedings of the 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)*. 2024:16965-16974. doi: 10.1109/CVPR52733.2024.01605Liao Z., Zhao Y., Shan X., et al. RT-DETRv4: Painlessly Furthering Real-Time Object Detection with Vision Foundation Models. arXiv preprint, 2025. doi: 10.48550/arXiv.2510.25257Liao Z, Zhao Y, Shan X, et al. RT-DETRv4: Painlessly Furthering Real-Time Object Detection with Vision Foundation Models. arXiv preprint. 2025. doi: 10.48550/arXiv.2510.25257Zhang Y., Sun P., Jiang Y., et al. ByteTrack: Multi-Object Tracking by Associating Every Detection Box // Computer Vision - ECCV 2022. Lecture Notes in Computer Science. Vol. 13682. Springer, Cham, 2022. P. 1-21. doi: 10.1007/978-3-031-20047-2_1Zhang Y, Sun P, Jiang Y, et al. ByteTrack: Multi-Object Tracking by Associating Every Detection Box. In: *Computer Vision - ECCV 2022*. Lecture Notes in Computer Science. Vol. 13682. Cham: Springer; 2022:1-21. doi: 10.1007/978-3-031-20047-2_1Aharon N., Orfaig R., Bobrovsky B.-Z. BoT-SORT: Robust Associations Multi-Pedestrian Tracking. arXiv preprint, 2022. doi: 10.48550/arXiv.2206.14651Aharon N, Orfaig R, Bobrovsky BZ. BoT-SORT: Robust Associations Multi-Pedestrian Tracking. arXiv preprint. 2022. doi: 10.48550/arXiv.2206.14651Ivanovska M., Kreft J., Štruc V., Perš J. Privacy-by-design AIoT vision for intelligent urban environments // Journal of Systems Architecture. 2025. Vol. 169. Article 103586. doi: 10.1016/j.sysarc.2025.103586 EDN: SJVYLQIvanovska M, Kreft J, Štruc V, Perš J. Privacy-by-design AIoT vision for intelligent urban environments. J Syst Archit. 2025;169:103586. doi: 10.1016/j.sysarc.2025.103586 EDN: SJVYLQKirillov A., Mintun E., Ravi N., et al. Segment Anything // Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). 2023. P. 4015-4026. doi: 10.1109/ICCV51070.2023.00371Kirillov A, Mintun E, Ravi N, et al. Segment Anything. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). 2023:4015-4026. doi: 10.1109/ICCV51070.2023.00371Ravi N., Gabeur V., Hu Y.-T., et al. SAM 2: Segment Anything in Images and Videos. arXiv preprint, 2024. doi: 10.48550/arXiv.2408.00714Ravi N, Gabeur V, Hu YT, et al. SAM 2: Segment Anything in Images and Videos. arXiv preprint. 2024. doi: 10.48550/arXiv.2408.00714Tardy H., Soilán M., Martín-Jiménez J.A., González-Aguilera D. Automatic Road Inventory Using a Low-Cost Mobile Mapping System and Based on a Semantic Segmentation Deep Learning Model // Remote Sensing. 2023. Vol. 15, No. 5. Article 1351. doi: 10.3390/rs15051351 EDN: UACFIZTardy H, Soilán M, Martín-Jiménez JA, González-Aguilera D. Automatic Road Inventory Using a Low-Cost Mobile Mapping System and Based on a Semantic Segmentation Deep Learning Model. Remote Sens. 2023;15(5):1351. doi: 10.3390/rs15051351 EDN: UACFIZLiu C., Xie M., Yuan C., et al. Training-free open-set 3D inventory of transportation infrastructure by combining point clouds and images // Automation in Construction. 2025. Vol. 178. Article 106377. doi: 10.1016/j.autcon.2025.106377 EDN: OMAPSALiu C, Xie M, Yuan C, et al. Training-free open-set 3D inventory of transportation infrastructure by combining point clouds and images. Autom Constr. 2025;178:106377. doi: 10.1016/j.autcon.2025.106377 EDN: OMAPSABoutayeb A., Lahsen-Cherif I., El Khadimi A. When Machine Learning Meets Geospatial Data: A Comprehensive GeoAI Review // IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. 2025. Vol. 18. P. 13135-13191. doi: 10.1109/JSTARS.2025.3568715 EDN: DXWMGQBoutayeb A, Lahsen-Cherif I, El Khadimi A. When Machine Learning Meets Geospatial Data: A Comprehensive GeoAI Review. IEEE J Sel Top Appl Earth Obs Remote Sens. 2025;18:13135-13191. doi: 10.1109/JSTARS.2025.3568715 EDN: DXWMGQPospelov P.I., Kotov A.A. The integrated client server complex of Geovideo for field survey of the heterogeneous data characterizing highways // Science Journal of Transportation. International Cooperation Issue. 2014. Special Issue No. 05. P. 75-81. EDN: ZCGXCXPospelov PI, Kotov AA. The integrated client server complex of Geovideo for field survey of the heterogeneous data characterizing highways. Sci J Transp. 2014;(Special Issue No. 05):75-81. EDN: ZCGXCXLi Y., Yu R., Shahabi C., Liu Y. Diffusion Convolutional Recurrent Neural Network: Data-Driven Traffic Forecasting // International Conference on Learning Representations (ICLR). 2018. Дата обращения: 15.06.26. Режим доступа: https://openreview.net/forum?id=SJiHXGWAZYu B., Yin H., Zhu Z. Spatio-Temporal Graph Convolutional Networks: A Deep Learning Framework for Traffic Forecasting // Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence (IJCAI-18). 2018. P. 3634-3640. doi: 10.24963/ijcai.2018/505Li Y, Yu R, Shahabi C, Liu Y. Diffusion Convolutional Recurrent Neural Network: Data-Driven Traffic Forecasting. In: International Conference on Learning Representations (ICLR). 2018. Accessed: 15.06.26. Available from: https://openreview.net/forum?id=SJiHXGWAZJiang W., Luo J. Graph Neural Network for Traffic Forecasting // ISPRS International Journal of Geo-Information. 2023. Vol. 12, No. 3. Article 100. doi: 10.3390/ijgi12030100 EDN: AIVOIFYu B, Yin H, Zhu Z. Spatio-Temporal Graph Convolutional Networks: A Deep Learning Framework for Traffic Forecasting. In: *Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence (IJCAI-18)*. 2018:3634-3640. doi: 10.24963/ijcai.2018/505Al-Huthaifi R., Li T., Al-Huda Z., et al. FedGODE: Secure traffic flow prediction based on federated learning and graph ordinary differential equation networks // Knowledge-Based Systems. 2024. Vol. 299. Article 112029. doi: 10.1016/j.knosys.2024.112029 EDN: TGQYKMJiang W, Luo J. Graph Neural Network for Traffic Forecasting. ISPRS Int J Geo-Inf. 2023;12(3):100. doi: 10.3390/ijgi12030100 EDN: AIVOIFChauhan S.S., Jain Y.K., Mannepalli P.K., Pandey A. 6G conditioned spatiotemporal graph neural networks for real time traffic flow prediction // Scientific Reports. 2026. Vol. 16. Article 3902. doi: 10.1038/s41598-025-32795-0 EDN: NMQDXGAl-Huthaifi R, Li T, Al-Huda Z, et al. FedGODE: Secure traffic flow prediction based on federated learning and graph ordinary differential equation networks. Knowl-Based Syst. 2024;299:112029. doi: 10.1016/j.knosys.2024.112029 EDN: TGQYKMLim B., Arık S.Ö., Loeff N., Pfister T. Temporal Fusion Transformers for interpretable multi-horizon time series forecasting // International Journal of Forecasting. 2021. Vol. 37, No. 4. P. 1748-1764. doi: 10.1016/j.ijforecast.2021.03.012 EDN: QOANABChauhan SS, Jain YK, Mannepalli PK, Pandey A. 6G conditioned spatiotemporal graph neural networks for real time traffic flow prediction. Sci Rep. 2026;16:3902. doi: 10.1038/s41598-025-32795-0 EDN: NMQDXGLim B, Arık SÖ, Loeff N, Pfister T. Temporal Fusion Transformers for interpretable multi-horizon time series forecasting. Int J Forecast. 2021;37(4):1748-1764. doi: 10.1016/j.ijforecast.2021.03.012 EDN: QOANAB