Lane Recognition on Country Roads

Most of the common lane recognition systems are designed to work on well structured roads and rely on the existence of markings. In this paper we present a lane recognition scheme for country roads. Our novel approach works even in the absence of markings. The parameter estimation is formulated as a maximum-a-posteriori estimation task fusing color, texture, and edges. The framework can easily be extended by additional features not considered here. The optimization is carried out by means of a particle filter. Efficient computation schemes allow running the system in video real-time using a standard PC. The proposed algorithm can cope with varying feature statistics. Practical tests prove the robustness on marked as well as unmarked roads.