Globalized probability based lane detection with non-unique B-spline model

In this paper, we proposed a robust lane detection method. This method uses the globalized probability of boundary (gPb) algorithm as boundary detector and non-unique B-spline (NUBS) as the road model. The gPb algorithm combines the local information, like brightness, color and texture features, with global information derived from spectral partitioning and is robust against shadow, and illumination varying in road boundaries detection. NUBS model is flexible enough to model the most of roads, include highways and rural ways, while taking account of the perspective effect. This method contains three main steps: first, we use gPb to detect the boundaries of the objects in the resized captured images, and then, take use of the parallel property of the road boundaries in ground images to find the knot sets of NUBS model for two sides of the road boundaries. Finally, we make the knots as initial boundaries and push them to the more accuracy boundaries in the canny edge map image using the snake algorithm.