Lane Detection for Intelligent Vehicles in Challenging Scenarios

Various image processing techniques and geometric models have been applied in vision based lane detection subsystems of intelligent vehicles and Advanced Driver Assistance Systems (ADAS). However, challenging conditions such as strong shadows, occlusions, eroded markings, high curvatures are ongoing issues in this topic. In this paper, a novel lane extraction method based on symmetrical local threshold is proposed with the hyperbola road model. Optimal model fit of lane distribution and hyperbola-pairs are estimated with RANdom SAmple Consensus (RANSAC) technique. The system computes the road geometry, vehicle position and direction both on urban road and country roads. Results have been presented on images with strong shadows, high curvature and eroded markings. Road and vehicle parameters computed on synthetic images are compared with ground truth information to show the accuracy of the proposed method.