Deep learning classification applied to traffic accidents prediction
Published in XLIII JORNADAS AUTOMÁTICA, 2022
In this paper, YOLOv4 neural networks are trained on a dataset, VisDrone, that is firstly validated through a custom-made graphic user interface. Several tests have been undertaken to compare these models to other previously existing models. The trained models are then used to detect and classify in real-time objects in a street to then be used by an algorithm to first track them, then calculate their trajectories and finally predict potential collisions between them.
Download here