General Road Detection From A Single Image

In this project, we implemented an algorithm that, given an arbitrary road image, delineates the possible path of travel. In contrast to urban roads, here the road can be unpaved, its boundaries unclear and its color distribution unknown. The algorithm operates in two main stages:
• Finding a vanishing point that aligns with the central (straight) part of the road.
• Segmentation of the road area based on the estimate of the vanishing point.
We find the vanishing point based on a voting function using a local voting area. For each pixel candidate to be vanishing point, the function sums up votes from the local area according to their Gabor filter response. Road segmentation is based on the assumption that there is a difference in the color texture between the road and other areas in the picture, so a sequence of pixels with similar orientations is calculated. The algorithm was implemented and tested with dozens of road images and proved to be effective in identifying road areas under challenging conditions.

General Road Detection From A Single Image
General Road Detection From A Single Image