Automatic Lane Parameters Extraction in Mobile Mapping Sequences

Technical prescriptions for road inventories specify that the road width must be measured with a tolerance of 10 cm and that variations larger than the same threshold should be spotted and recorded. In the framework of software development for Mobile Mapping, underway at the Department of Civil Engineering of Parma University, a semi-automatic method for the extraction, the tracking along an image sequence and the measurement of the distance between lane markings was developed. The input data is a sequence of stereo images with known interior and exterior orientation. The program starts by interactively defining a ROI for each lane to be extracted; both images are then rectified with respect to a common object plane; by edge detection the markings extracted in each image pair; edge pixel are selected and interpolated with a straight line and the distance between these lines measured. From their differences the true distance and the height difference due to the cross slope are computed; the measurement is repeated at specified intervals and recorded. By predicting the position of the markings in the next image pair, the sequence is scanned as long as the markings can be reliably extracted. When, for a specified number of successive pairs, no marking is found, the operator is requested to restart the process. The first trials on artificial and real image sequences have shown accuracy well within the specifications and the robustness of the extraction.