Intelligent Recognition of Emergency Vehicles in Congested Traffic using Computer Vision

Timely identification of emergency vehicles in congested traffic environments is critical for improving response times and enhancing public safety. This study presents an intelligent computer vision-based system for real-time emergency vehicle detection and instance segmentation using a YOLOv12-based deep learning architecture. A hybrid dataset was constructed by integrating a publicly available emergency vehicle dataset with locally collected Philippine traffic images to capture region-specific visual characteristics. The model was trained using optimized data augmentation and hyperparameter configurations and evaluated through iterative experimentation. Experimental results demonstrate strong performance, achieving an overall precision of 0.948, recall of 0.890, mAP@0.50 of 0.951, and mAP@0.50:0.95 of 0.871. Class-wise analysis shows particularly high detection accuracy for ambulances and fire engines, while qualitative and confusion matrix analyses confirm robust generalization on unseen data with limited misclassification primarily involving non-emergency vehicles. The model was further deployed through a Gradio-based web interface, enabling real-time interactive testing with visualized detection outputs. The results indicate that the proposed system is effective, stable, and suitable for real-world emergency vehicle recognition applications within intelligent transportation systems.