In developing nations, traffic congestion delays emergency vehicles like ambulances, causing precious time and life-threatening delays. Pedestrian safety is also a principal issue in urban areas with rising population and vehicular concentrations. This paper proposes an intelligent traffic management system integrating vision-based YOLOv8 object detection and audio-based siren detection for real-time ambulance identification and pedestrian crowd density assessment. The system employs YOLOv8 neural networks for ambulance and pedestrian detection, complemented by Mel-frequency cepstral coefficients (MFCC) and CNN-based audio processing. A novel adaptive decision fusion framework dynamically integrates multi-modal data based on environmental context. Experimental validation demonstrates 92.3% precision for ambulance detection and 95.1% for pedstr riandetection, with 35% false positive reduction compared to single-modality approaches. Real-world deployment shows 28.3% improvement in emergency response times with 99.2% system availability.