UGHYT-PLUS: YOLOV11-ENHANCED FEDERATED RL FRAMEWORK FOR ULTRA-PRECISE SPEED ESTIMATION, ANOMALY DETECTION, AND ADAPTIVE TRAFFIC CONTROL IN OCCLUSION-HEAVY URBAN SURVEILLANCE
DOI:
https://doi.org/10.48047/aj3x9f15Palabras clave:
YOLOv11, federated reinforcement learning, self-supervised anomaly detection, evidential deep learning, traffic speed estimation, edge computing, intelligent transportation systems, occlusion-robust surveillanceResumen
Urban traffic in India claims over 150,000 lives yearly, with overspeeding and anomalies exacerbating fatalities in occlusion-heavy, low-resolution CCTV environments like Kolkata and Bhopal. Building on the UGHYT framework, this paper introduces UGHYT-Plus—a YOLOv11-enhanced federated reinforcement learning (RL) system fusing self-supervised anomaly detection, evidential speed regression, and adaptive signal control. UGHYT-Plus achieves 96.2% mAP, 1.8 km/h RMSE, and 96% anomaly recall at 25 ms latency on expanded MITS/BDD100K datasets plus custom 75-hour Kolkata CCTV (80% occlusion, 480p, 8-20 FPS). Innovations include contrastive pretraining for incident detection (accidents, emergency vehicles), multi-modal radar fusion for sub-2 km/h precision, and privacy-preserving federated RL optimizing signals to cut congestion by 28%. Outperforming YOLOv10 baselines by 35% in accuracy and 30% in speed, UGHYT-Plus enables scalable 5G-IoT deployment across 100+ feeds on 20W Jetson nodes, paving the way for safer smart cities.
Descargas
Referencias
J. Kim et al., “Real-time vehicle counting and speed estimation using deep learning,” J. Adv. Transp., 2021.
W. Luo et al., “Multi-object vehicle tracking in urban scenes,” IEEE Access, vol. 9, pp. 45806–45819, 2021.
S. T. N. Tran et al., “Real-time vehicle counting and velocity estimation,” Sustain. Cities Soc., vol. 102, 105234, 2024.
S. Li et al., “YOLOv8-DeepSORT vehicle speed detection,” arXiv:2406.07710, 2024.
R. Wang et al., “Occlusion-robust tracking with YOLOv9 and BoT-SORT,” Procedia Comput. Sci., vol. 235, pp. 1234–1242, 2024.
Kumar et al., “Multi-stage deep learning for vehicle detection and speed estimation,” PMC, 2025.
Ultralytics, “YOLOv11 traffic and edge inference APIs,” Tech. Docs., 2026.
M. Tiwari and V. Sakalle, “UGHYT: Uncertainty-guided traffic speed monitoring,” LNCT Univ., 2026.
W. Luo et al., “Vision–mmWave radar fusion for vehicle tracking,” IEEE Access, 2021.
Z. Chen et al., “Self-supervised anomaly detection in traffic,” Sci. Rep., 2025.
K. Singh and A. Roy, “Contrastive anomaly detection for traffic incidents,” IEEE Trans. ITS, 2025.
Verma et al., “Federated learning for intelligent transportation systems: A survey,” arXiv:2403.07444, 2024.
Z. Chen et al., “Vision-supervised federated learning for traffic monitoring,” arXiv:2506.19023, 2025.
M. Chatterjee et al., “TrafficEZ: Edge-based real-time surveillance,” MJST, 2025.
S. Patel, “YOLO-based traffic speed estimation in Indian cities,” J. Intell. Syst., 2025.
Descargas
Publicado
Número
Sección
Licencia
Derechos de autor 2024 Mohit Tiwari, Dr. Vikas Sakalle (Author)
Esta obra está bajo una licencia internacional Creative Commons Atribución 4.0.
Usted es libre de:
- Compartir — copiar y redistribuir el material en cualquier medio o formato para cualquier propósito, incluso comercialmente.
- Adaptar — remezclar, transformar y construir a partir del material para cualquier propósito, incluso comercialmente.
- La licenciante no puede revocar estas libertades en tanto usted siga los términos de la licencia
Bajo los siguientes términos:
- Atribución — Usted debe dar crédito de manera adecuada , brindar un enlace a la licencia, e indicar si se han realizado cambios . Puede hacerlo en cualquier forma razonable, pero no de forma tal que sugiera que usted o su uso tienen el apoyo de la licenciante.
- No hay restricciones adicionales — No puede aplicar términos legales ni medidas tecnológicas que restrinjan legalmente a otras a hacer cualquier uso permitido por la licencia.
