Automatic road object extraction from Mobile Mapping Systems

Mobile Mapping Systems (MMSs) often represent the best choice to provide an accurate 3D modeling of the environment, especially in urban streets where the aerial/satellite surveys do not provide accurate data. MMSs are equipped with many kinds of sensors, and, in particular, laser scanners that allow 2D/3D environment modeling from very dense point clouds. Usually an operator manually explores the point cloud to discover and mark a particular feature of interest (e.g., road line, cross-walk). Obviously this procedure is tedious and expensive. One of the greater challenges is to automatically extract objects/features from co-registered data coming from LiDAR, optical and positioning sensors. This paper presents an automatic feature/object approach to extract and then to georeference with high accuracy/precision horizontal road signs, mainly lanes and crosswalks. The proposed approach exploits image processing techniques and methods for the 3D to 2D re-projection of data. The results obtained demonstrate that is possible to achieve accuracy and precision in the range of one centimeter.