Road detection, which brings a visual perceptive ability to
vehicles, is essential to build driver assistance systems. To
help detect lane markings in challenging scenarios, one-time
calibration of inverse perspective mapping (IPM) param-
eters is employed to build a bird’s eye view of the road
image. We propose an automatic IPM method based on road
boundaries called BIRD (Boundary-based IPM for Road
Detection), avoiding common problems of fixed IPM. Fur-
thermore, integrating top-down and bottom-up attention, an
illumination-robust lane marking detection approach using
BIRD is proposed.

Source code is available at https://github.com/baidut/OpenVehicleVision