The recognition and tracking of lanes is a fundamental task for the guidance of autonomous vehicles in urban scenarios. Lane detection is the problem of determining the position of the vehicle within the lane and the road course ahead. Lanes are typically defined by painted markings on the road that define the course along which the vehicle is able to advance. Vision systems are usually used to extract the road markings and estimate the state of the vehicle with respect to the lanes. The extraction of the lanes from the visual data is carried out by defining the set of features that best detect and discriminate the lane markings. The proper detection of the lane is performed by fitting the features to a road model. The model is also usually tracked over time in order to reduce uncertainty and to provide an a-priori estimate of the lanes for the next system cycle.
The detection of intersections is obtained in a similar way. Features are extracted from the environment and matched to a database of features previously created. The database consists of a list of intersections with the descriptors found in each intersection. The descriptors are a set of discriminant features that can be uniquely assigned to an intersection. As the vehicle drive in its environment, the acquired features are compared to those in the data base. The detection occurs when a significant amount of features are found for a given intersection. This method is also known more generally as place recognition.