There are other alternatives to the classic techniques based on Deep Learning. These approaches have been implemented in two ways:
-LaneNet ROS package;
The former consists of an encoder-decoder stage, binary semantic segmentation stage and instance semanticsegmentation for a near real time lane detection task. The latter is based on a Fast AI lesson (tutorial) and it is not a ROS package but presents reliable results with small Dataset to trian the network. The type of learning technique that is going to be implemented is the semantic segmentation applied ina real road scenario.
The results obtained were presented on the two videos below (to the LaneNet ROS package and the UNet Model, respectively):