A New Color-Based Lane Detection Via Gaussian Radial Basis Function Networks

Lane detection plays a central role in intelligent transportation systems. While edge detection on intensity images has gained much popularity in the past, it usually results in noisy binary images. Most noticeably, the color information of the scene that may provide an important cue for lane detection has not been genuinely considered. In this paper, we propose a novel color-based lane detection system. Although color-based schemes have their fair share of issues, including varying illumination conditions, by relying on a lane mark color predictor obtained from an offline supervised training of Gaussian radial basis function (GRBF) networks, such issues can be appropriately overcome. Experimental results have demonstrated that the proposed approach, in contrast to predominantly edge-based approaches, can effectively eliminate erroneous edges that do not belong to the lane marks in well-structured scenes.