An intelligent YOLO and CNN-BiGRU framework for road infrastructure based anomaly assessment

Area of Science:

• Intelligent Transportation Systems
• Artificial Intelligence in Civil Engineering
• Deep Learning for Infrastructure Monitoring

Background:

• Traditional road inspection methods are inefficient, hazardous, and time-consuming.
• There is a need for automated, real-time solutions for intelligent road infrastructure management.
• Deep learning offers a transformative approach to overcome these limitations.

Purpose of the Study:

• To develop a novel, real-time monitoring framework for intelligent road infrastructure management.
• To integrate object detection (YOLOv11) and temporal severity prediction (CNN-BiGRU) within a digital twin (DT) environment.
• To address critical road conditions like potholes, cracks, obscured markings, and snow-covered surfaces.

Main Methods:

• Utilized YOLOv11 for precise object detection and CNN-BiGRU for temporal severity prediction.
• Employed a digital twin (DT) driven simulation environment for infrastructure context modeling.
• Evaluated the hybrid system on the LiRA-CD dataset (30,000+ instances) with a 70:15:15 split.
• Benchmarked performance on high-end hardware (Intel i7, NVIDIA RTX 3090) with TensorRT acceleration.

Main Results:

• Achieved real-time inference at 105 FPS with a per-frame latency of 9.5 ms.
• Attained high accuracy metrics: mAP@0.5 of 96.92%, mAP@[0.5:0.95] of 90.74%, and AUROC of 0.942.
• Demonstrated strong regression performance for severity prediction (R²=0.77, AAE=0.31, ASE=0.34), indicating robustness.

Conclusions:

• The integrated framework provides a scalable and efficient solution for proactive road infrastructure monitoring.
• This hybrid system sets a new benchmark for smart transportation systems through real-time, data-driven analysis.
• The combination of DT, YOLOv11, and CNN-BiGRU enhances adaptive infrastructure management capabilities.