Towards Efficient Learning Using Double-Layered Federation Based on Traffic Density for Internet of Vehicles.