An AI-driven IoT-enabled intelligent transportation framework for emergency response in smart cities

Abstract Delayed ambulance transit through congested urban intersections critically reduces patient survival rates in emergency medical situations. For cardiac arrest, survival probability decreases by approximately 7–10% per minute without intervention, making rapid emergency vehicle clearance a life-critical operational requirement. Existing approaches rely on single-modality vehicle detection, fragmented inter-agency communication, or costly vehicle-side hardware, each imposing substantial limitations that hinder city-scale deployment. This paper presents an AI-Driven IoT-Enabled Intelligent Transportation Framework for ambulance prioritisation and emergency coordination in smart cities, comprising four integrated subsystems. A multimodal deep learning pipeline combining a YOLOv11-Small visual detector and a Convolutional Neural Network (CNN) siren classifier through a confidence-based adaptive weighted fusion algorithm achieves 98.4% overall detection accuracy, 97.8% F1-score, and 21 ms inference latency, outperforming a retrained YOLOv8 baseline and prior multimodal systems evaluated under identical dataset and training conditions. Upon confirmed detection, a Raspberry Pi IoT edge device actuates physical traffic signal relay hardware; the detection-pipeline-to-MQTT-dispatch latency measures 55–90 ms, with an additional MQTT-to-relay-actuation time of 120–180 ms yielding a total physical signal actuation latency of 175–270 ms — well within the 200–500 ms phase transition timescales of standard adaptive traffic controllers. A cross-platform tri-role Mobile Emergency Medical Services (MEMS) application connects ambulance staff, police officers, and hospital personnel with JWT-secured role-based access control and a human-in-the-loop signal override capability. An in-ambulance YOLOv11 patient condition monitoring module is presented as a proof-of-concept implementation detecting four critical clinical indicators from closed-circuit camera video and delivering structured pre-arrival alerts to the hospital dashboard. An ablation study confirms the contribution of each subsystem, and evaluation under nine condition combinations (three traffic densities × three illumination levels) demonstrates fusion accuracy above 97.1% across all tested scenarios.