In the last years many research groups and companies have worked on the development of self-driving cars. Among the first layers of automation there is the lane keeping functionality. This thesis introduces two lane keeping control algorithms and a computer vision algorithm for lane identification through a camera mounted on the vehicle. In the preliminary part of the thesis algorithms for tracking the center of the lane have been developed through PID control and optimal control techniques such as non-linear MPC. In this phase Matlab and Simulink were widely used. Subsequently, urban environments with intersections, vehicles at close range and high presence of vertical and horizontal signs have been addressed through convolutional neural networks and the NVIDIA model but no training has been done for lack of computational power.