A DL-based estimation probability approach for VRU collision avoidance

Assisted/autonomous driving is nowadays a vivid sector for both research and industry, thanks to the advances made using artificial intelligence and the pervasive and fast communication achieved by the Fifth generation (5G) of cellular networks. A fully connected environment where traveling on public roads is done with limited (or without) human intervention may increase road safety.

In this work, we present a system to detect possible collisions among vehicles and between pedestrians and vehicles with the final aim of reduce traffic accidents. Our proposal is based on a trajectory prediction algorithm plus a method to estimate the collision probability. Deep learning and Monte Carlo algorithms are used, respectively. The promising results open future research extensions.


  • Raúl Parada, CTTC/CERCA, Madrid, Spain.
  • Rafael Corvillo, UOC, Barcelona, Spain.
  • Paolo Dini, CTTC/CERCA, Madrid, Spain.


You can also read this article in Zenodo.