Università degli studi di Modena e Reggio Emilia

This paper provides two novel contributions to vehicular cooperative perception. Firstly, it puts forth an approach to generate the actual perception messages broadcasted by connected autonomous vehicles. Relying on data gathered by autonomous vehicles and originally collected for computer vision purposes, it produces perception messages in accordance with the standard ETSI rules. The statistical properties of the messages are determined, showing that their size is remarkably affected by the driving scenario and the policy adopted to discern when an object is seen by the vehicle, and to a lesser extent by the selection of the message generation frequency. Secondly, the paper proposes a generative model to synthetically replicate the sequences of perception messages. The ability of the model to successfully capture the characteristics and the temporal correlation of the real data is demonstrated in a reference scenario. The model adoption is promising in largescale numerical simulations, where the perception messages of many vehicles have to be faithfully reproduced.