Robust road modeling based on a hierarchical bipartite graph

Driver assistance systems based on video processing deliver a number of warnings to the driver, such as lane departure, lane invasion by other vehicles, collision prediction, etc. This have been a field of intense research for many years, providing solutions based on road models where vehicles are afterwards detected and tracked. Robustness is essential in this field of road safety where outliers represent one of the major problems for road modeling. The motivation of this work is to provide a robust and, at the same time, flexible road model which identifies a variable number of lanes, their widths, the curvature of the road and the position of the vehicle in its lane. The major advantage of this model is that the system gives confidence measures for each lane, determining which lanes are actually present and which not. The model is structured as a hierarchical bipartite graph which simplifies information management, reduces sub-module dependencies and classifies elements of the road in different levels. At each level different strategies are applied, following four overall steps: measurement, estimation, evaluation and extrapolation, which lead to enhanced road model accuracy, reliability and flexibility. Several experimental results are provided, showing the robustness of the system, its stability and accurate results for large test paths.