Exploiting infrastructure sensors for advanced vehicle manoeuvring assistance

In the last decade, there has been a huge increase in studies regarding autonomous vehicles and their use in real scenarios. Autonomous vehicles are now becoming safe and reliable also in everyday traffic, thanks also to a great number of sensors used to perceive the world around them. However, there are situations in which those sensors may not be so effective, mainly if something is blocking their view. Smart infrastructure can help connected vehicles, providing information about the road status even before the vehicle is able to sense it. In this thesis, it has been developed an entire software pipeline that extracts information from roadside sensors data and it uses this information to suggest to connected vehicles the manoeuvres to perform at an intersection. The roadside unit (RSU) is equipped with a camera and a LiDAR; data from these sensors are processed to identify all road actors, estimate their positions and predict their future trajectories. The software architecture is composed of a set of modules that communicate with each other through a broker; this architectural choice provides high flexibility to the system, allowing to update or add modules easily. The main modules of the system are the ones that identify road actors from the video, merge this information with the data from the LiDAR to compute the actors' position and volume, and use this position to compute their future trajectories. To ensure the consistency of the data extracted from the camera and the LiDAR when they are merged, this thesis also deals with the calibration and synchronization of these sensors. The computation will not be performed on the roadside unit itself, but instead in a server located at the edge of the mobile network (Edge Server), that will receive sensors data flows from one or more RSUs.