Robust lane detection based on gradient direction

A robust and effective method to detect lane in the images captured with a vehicle-mounted monocular camera in challenging environments is proposed in this paper. In the newly proposed approach, the gradient direction (GD) feature and the lane boundaries projection model are used. Using GD feature and GD Gaussian distribution with the likelihood function, the lane detection is performed by employing maximum a posteriori (MAP) estimation with prior knowledge. Afterwards, the model parameter values are estimated, with which the lane geometric structure (such as the lane curvature and change rate), the host vehicle position and heading direction in the lane can be also calculated. The experimental results show that the method works more robustly and accurately in various situations with the broken and worn lane markings, the curved lane, the messy shadows, the sun glare, the occlusion of other vehicles, the dusky light in the evening, etc.