Aggregate road surface based environment representation using Digital Elevation Maps

A Multivariate Gaussian probability distribution based solution is proposed for constructing Digital Elevation Maps. The constructed Digital Elevation Map is used for estimating the road surface and classification. A novel approach of representing the road surface is also used to model the more complicated scenarios of road surfaces. Instead of a single planar or quadratic road surface, several consistent quadratic road surfaces are found, forming an aggregate road surface. The main quadratic road surface starting with a patch in front of the vehicle is fitted as a plane using a modified version of RANSAC, called CC-RANSAC (Connected Component Random Sample Consensus). A region growing process is employed to obtain the quadratic representation of the main road surface. The growing process uses as stopping criteria the homogeneity of neighboring cells with respect to the current road surface, taking into account the 3D uncertainty model. The secondary road surfaces are obtained starting with the points from the boundary of the main road surface. The corresponding regions are grown as long as the neighboring cells are connected based on their height, performing set union when multiple patches meet. The consistent secondary road surfaces are kept and an aggregate road surface is obtained by performing a segmentation of the environment. A correspondence is made between road surfaces and environment areas. The obtained aggregate road surface is robust and works well in scenarios where multiple roads intersect.