Recovering a comprehensive road appearance mosaic from video

We describe a system that employs a single calibrated camera mounted on a moving vehicle to produce a road appearance map as a comprehensive mosaic of individual orthogonal views. The system first transforms the current image of the road acquired from a driver's perspective into the orthogonal view by inverse perspective mapping. Consequently, the orthogonal image is aligned with previously recovered parts of the mosaic by an exhaustive search optimization technique. Experiments have been performed on videos taken along public roads through Croatian countryside and small cities. The obtained results are provided and discussed.