Multimodal Deep Learning Framework for Intelligent Traffic Signal Control with Emergency Vehicle Prioritization

Abstract

Urban traffic congestion and delayed emergency response times represent critical challenges in modern smart cities. This comprehensive review examines recent advances (2024-2025) in intelligent traffic management systems that integrate deep learning-based vehicle detection with adaptive signal control mechanisms, specifically focusing on emergency vehicle prioritization. We analyze 30 state-of-the-art systems that leverage YOLOv9 architectures, multimodal fusion techniques, and edge computing platforms to achieve real-time traffic optimization. Key findings reveal a paradigm shift toward attention-enhanced detection models (YOLOv9+CBAM), audio-visual fusion for robust emergency vehicle identification, and edge deployment on resource-constrained hardware (Raspberry Pi, Jetson platforms). Systems employing multimodal confirmation mechanisms demonstrate superior reliability, with reported accuracies exceeding 96% and response time reductions of up to 35%. However, standardized benchmarking for false positive rates remains limited. This review synthesizes architectural innovations, prioritization strategies, and performance characteristics to provide a comprehensive framework for researchers and practitioners developing next-generation intelligent transportation systems. We identify critical research gaps and propose future directions toward more reliable, scalable, and context-aware traffic management solutions.