Mosaic-Mixed Attention-Based Unexpected Traffic Scene Classification

Abstract

With the rapid advancements in artificial intelligence and smart mobility technologies, traffic monitoring systems are evolving quickly. Among these systems, in-vehicle monitoring systems using Object Detection (OD) algorithms are gaining attention for identifying traffic participants in distress. However, current OD algorithms often underperform in complex or unexpected traffic scenarios, such as accidents. In this study, we propose a novel scene classification network that integrates OD with a cross-attention mechanism By leveraging spatial mosaic and mixed attention mechanisms, the network emphasizes spatial relationships and inter-channel correlations, significantly enhancing accuracy in identifying critical traffic events. The detailed evaluation demonstrates improved efficiency and accuracy, underscoring the potential of the system for future traffic incident classification.