This is a solution to the 2019 RVSS workshop,finished by me and @Josh Newans,engineer who works in Newcastle.
We use a classification method to train a shallow CNN model and deploy it on a mobile-robot, which enables the robot to drive along the white-green lane with gaps. Before that, we manually drive the robot along the lane and collect for more than 600 images. We use random blurring to apply the data augmantation. Flip might be a better way but we didn't try it(remember to simultaneously flip the image and label)
##Opencv3
##Pytorch
##torchvision
Thanks Feras for organizaing such a nice workshop ! The original provided code and background are here