Traffic Incident Detection in Urban Roads Based on Hybrid 1D-CNN and Residual Transformer

Abstract

One of the most essential tasks for guaranteeing safety and operational efficiency in urban road traffic management is incident detection in real-time scenarios. In this paper, we propose a new hybrid framework that focuses on unidimensional convolutional neural networks (1D-CNN) for spatial feature extraction, residual transformers for temporal data modeling, and extreme gradient boosting (XGBoost) for efficient incident classification. This combination of strengths and performances, such as the robustness of 1D-CNN in spatial analysis and residual transformers in the capture of long-range dependencies in the case of temporal data, ensures robust feature extraction for the proposed model. The proposed framework offers significant competitive advantages and high precision, as demonstrated by experimental results.