Lane confidence fusion for visual occupancy estimation

Detection and tracking of road lanes is vital to a wide range of driver assistance systems. Confidence measures on tracked lanes are necessary to create reliable assistance systems. On curvy, multi-lane roads, adaptive cruise control systems could greatly benefit from information of lane location and curvature to identify the leading vehicle.We present a novel approach to estimating unobstructed lane length using visual lane confidence measures. The gradient between individually calculated lane confidence segments of a lane model provided by a lane tracker is used to estimate the distance to leading cars or obstacles on lanes. To achieve this goal, various confidence measures are defined, and a confidence fusion method based on Dempster Shafer is presented.