Mapping from black and white images

edit

If a few conditions can be assumed then this can certainly be done. If you can safely assume that the ground is always flat, the robot is always placed horizontally on the ground at the same height and the mono-camera is fixed on the robot body then you should be able to do this. There will be some edge cases such as someone with yellow shoes on walking infront of the robot and such like, but overall it should be okay.

Using the assumptions above, and segmenting out the yellow lines from the camera image, you can use the calibrated lens model of the camera and the assumption of a flat surface to project these yellow lines into an orthographic (top down) image of the yellow boundaries. A point cloud would actually be a useful type to use here because it could be fed into existing SLAM nodes. Also if there were long straight yellow roads then localisation using the camera alone would be impossible, so you would need to combine this with wheel odometry too.