Exploitation of road signalling for localization refinement of autonomous vehicles

In this paper we present the development of a framework for vehicle localization refinement based on the detection of traffic road signs. Leveraging on the detection capabilities of deep neural networks, the proposed architecture relies on the recognition of traffic signs to navigate a graph, in order to spatially localize the vehicle on the road. Knowing the internal camera parameters and the size of the road signals it is possible to refine the localization accuracy and propagate the information over time thanks to the application of a Kalman filter. The implemented solution demonstrates that the trajectory of the vehicle is more accurate, reducing the error when comparing the standard GPS information and the RTK positioning system.