Intelligent road lane mark extraction using a Mobile Mapping System

Abstract

During the last years, road landmark in- ventory has raised increasing interest in different areas: the maintenance of transport infrastructures, road 3d modelling, GIS applications, etc. The lane mark detection is posed as a two-class classification problem over a highly class imbalanced dataset. To cope with this imbalance we have applied Active Learning approaches. This Thesis has been divided into two main com- putational parts. In the first part, we have evaluated different Machine Learning approaches using panoramic images, obtained from image sensor, such as Random Forest (RF) and ensembles of Extreme Learning Machines (V-ELM), obtaining satisfactory results in the detection of road continuous lane marks. In the second part of the Thesis, we have applied a Random Forest algorithm to a LiDAR point cloud, obtaining a georeferenced road horizontal signs classification. We have not only identified continuous lines, but also, we have been able to identify every horizontal lane mark detected by the LiDAR sensor.