Robust multi-lane detection method based on semantic discrimination

Lane detection is the most basic part of automatic driving, driver assistance and violation-detection systems. A lane detection method based on semantic discrimination is proposed to address the problems of diversity in illumination, occlusion and lack of prior knowledge about lane markings. Two symmetric ridge operators are designed to improve the precision of candidate-pixel extraction. The double-constrained random sample consensus method (DC-RANSAC) introduces colour and geometric constraints into the lane fitting to reduce the number of iterations and improve the accuracy of the hypothetical model. Classifiers are added to further validate the hypothetical model and identify its underlying semantics. The proposed method was evaluated using two different data sets with various scenarios, including unclear lane markings, dense traffic, occlusion of vehicles, complex shadows, road surface markings, poor lighting conditions, and unknown number of lane markings. The detailed evaluations show that the detection rate of the proposed method is comparable with that of existing state-of-the-art lane detection methods, whereas the precision rate is much higher. Moreover, the experiments prove the reliability of the proposed algorithm in lane marking semantic recognition.