An Intelligent Traffic Conflict Prediction Using Deep Learning with Long-Term Evolution Access Data

Urban road safety management relies heavily on the ability to foresee conflicts at signalized junctions in realtime. By combining lane-level information with feature interactions and utilizing video data realtime recognition technologies, this study created a framework for realtime conflict anticpation at signalized crossings. First, we extract and process video data in real-time. Then, we build a realtime traffic conflict anticpation model based on Deep Cross Networks (DCNs). Finally, we use Shapley-AdditiveExPlanations(SHAP) to analyze the interpretability of conflictdriven factors. The initial step is to develop a reliable system for automatically extracting vehicle trajectories in order to identify conflict frequencies and dynamic traffic factors in real time. The second step is to build a DCN model that can represent the interrelationships and conflicts among the many dynamic traffic characteristics. Stage three of SHAP involves investigating how various dynamic traffic characteristics affect traffic conflicts. We test the model's interpretability and predictive power with video data from intersections. Methods for data-driven road safety management can find a safety measurement standard in the suggested framework.