A reliable road lane detector approach combining two vision-based algorithms

We present a new approach for detecting road lane with reliability using vision-based techniques. The single used sensor is an on-board frontal monocular monochromic camera. The novelty is to introduce redundancy by using several (two in this paper) independent algorithms. By combining their outputs together, we obtain more reliable results as well as a confidence value about the reliability of the results. When the confidence value is low, an initialization of the algorithms is performed to avoid false detection. The first algorithm computes longitudinal-coherent results, whereas the second algorithm computes lateral-coherent results. First, each algorithm is presented and then the way in which their outputs are combined is described. We also discuss how the two algorithms interact together. Then, experiments in a real-time context underline the improvements prodded by this approach and show that results are relevant.