SMARTACCIDENTAI: DEEP ENSEMBLE-BASED REAL-TIME DETECTION FRAMEWORK FOR SMART CITY TRANSPORTATION SYSTEMS

Abstract

Rapid urbanization and the exponential rise in vehicle density have made traffic accident detection an increasingly critical aspect of modern smart city infrastructure. Traditional surveillance and manual monitoring systems are inefficient for real-time accident response, often leading to delayed emergency assistance and preventable fatalities. To address these challenges, this paper proposes SmartAccidentAI, a Deep Learning Ensemble Framework designed for intelligent, real-time traffic accident detection. The model integrates multiple deep learning architectures—such as Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and Vision Transformers (ViTs)—to analyze live traffic video streams and sensor data for accurate event classification. By combining spatial and temporal features, SmartAccidentAI effectively distinguishes between normal and abnormal vehicular motion patterns, enabling early accident detection and automated alerts to emergency response units. Experimental results show that the ensemble approach achieves superior precision and recall compared to individual models. This work contributes to the advancement of smart city transportation safety, offering a scalable, accurate, and adaptive framework that can significantly reduce response times and improve urban mobility resilience.