Robust real-time lane detection based on lane mark segment features and general a priori knowledge

Lane detection plays an important role in vision based intelligent vehicle systems. A new lane detection method based on lane mark segment features and general a priori knowledge is proposed in this paper. Instead of detecting each feature point separately from limited local view, a lane mark segment detection method is designed for detecting each lane mark segment on the whole. Some a priori knowledge which is quite general for real traffic scenarios is used in the lane mark segment detection method as well as in the part of model fitting. The tracking process which ensures detection stability and robustness is carried out in the framework of particle filtering. The performance of the proposed method has been demonstrated based on the test on thousands of road images; these road images include scenarios with many kinds of uncertainties such as variation of lighting condition, existence of leading vehicles etc. The research direction for further improvements is also discussed.