Realistic Cooperative Perception for Connected and Automated Vehicles: A Simulation Review

This paper investigates realistic perception in vehicular simulations, demonstrating and comparing two different methods used mainly by the community. To render the photorealistic environment and generate RGB camera images, we use the open source simulator CARLA-SUMO co-simulation designed for automated driving research that allows us to generate realistic perception data. We show that the detection accuracy differs depending on vehicle layout, proportion of connected vehicles, weather conditions, and distance between the sensing vehicle and the target vehicle, which can also be realistically modelled in vehicular simulation. We show that a realistic perception is essential in vehicular simulations and can not be neglected or oversimplified to ensure a reliable Cooperative Perception Solution (CPS). Our experimental results show that a vision-based optical sensor is able to detect more vehicles than a grid-based projection sensor in cooperative perception scenarios.