A 360-degree Multi-camera System for Blue Emergency Light Detection Using Color Attention RT-DETR and the ABLDataset

A research team from Universitat Politècnica de València (UPV) has published a novel computer vision system designed to give vehicles a 360-degree awareness of approaching emergency services. Their paper, 'A 360-degree Multi-camera System for Blue Emergency Light Detection Using Color Attention RT-DETR and the ABLDataset,' presents a hardware and software solution using four fisheye cameras and a custom AI model. The core innovation is the integration of a specialized 'color attention block' into the RT-DETR (Real-Time Detection Transformer) architecture, which was selected after outperforming other models like YOLO v10 and Faster R-CNN. This enhancement allows the model to focus specifically on the critical blue light signature, a key feature for reliable detection amidst visual clutter.

The system was trained and tested on the team's own ABLDataset, a curated collection of European emergency vehicle images under diverse conditions. The results are significant: 94.7% accuracy and 94.1% recall on the test set, with successful real-world detections at distances up to 70 meters. Beyond simple detection, the calibrated camera setup uses geometric transformations to estimate the azimuthal location and approach angle of the emergency vehicle relative to the host car. The work is explicitly framed as a component for future multimodal ADAS, where this visual data would be fused with acoustic detection of sirens. This research provides a concrete, high-performance building block for next-generation safety systems that could automatically alert drivers or guide autonomous vehicles to yield effectively.