Detection of discontinuities of road curvature change by GLR

A new approach for spatial segmentation of a curved road for visual road recognition in far look-ahead distance, implemented in the autonomous road vehicle VaMP (a passenger car), is described. Based on the recognition of bright lane markings and the edges formed at the road borders, an estimation process performs an update of a state vector, which describes spatial road shape and vehicle orientation relative to the road. Kalman filter techniques for time-discrete measurements are applied to obtain an optimal estimate of the state vector at each time frame t_k. By extending the Kalman filter method for space-discrete measurements at a fixed time frame, the modification of the road curvature along the look-ahead axis can be recursively estimated. By applying the generalized likelihood ratio (GLR) approach, a segmentation of the road into segments with constant curvature change parameter can be obtained