SMARTROAD: A DEEP LEARNING-BASED ROAD EVENT AWARENESS DATASET FOR INTELLIGENT AUTONOMOUS DRIVING

Abstract

The rapid evolution of autonomous driving technologies has made the availability of high-quality, context-rich datasets crucial for improving road event understanding and situational awareness. This paper introduces SmartROAD, a novel deep learning-based road event awareness framework designed to enhance perception and decision-making in self-driving systems. The system utilizes advanced computer vision algorithms and multimodal sensor data to classify, localize, and predict real-time road events such as lane changes, pedestrian crossings, traffic signal violations, and accident risks. Unlike conventional datasets that focus on limited scenarios, SmartROAD provides dynamic annotations and contextual metadata, improving model generalization in diverse traffic environments. Experimental results demonstrate that integrating SmartROAD with deep learning models such as CNN-LSTM and Transformer architectures leads to significant improvements in event detection accuracy and driving safety metrics