Thursday, October 15, 1998
In this work we consider a problem arising in intelligent
vehicles and highway systems (IVHS), specifically in
automatic road following, collision avoidance, and maneuver
control. A millimeter-wave radar is placed on the front of
the vehicle and an image of the radar backscatter of the
terrain is acquired in polar coordinates. The radar sensor
acquires the scene in polar coordinates.
The obtained radar image contains both road and offroad scatter components and we present a method to segment them from each other. The variety and intesity of the offroad scatter components, the requirement of reliability under changing conditions (weather, visibility, road surface etc.), make it necessary to pertinently use the available constraints about the road geometry. These constraints are naturally formulated in the cartesian cartographic domain. Following [Lakshmanan96,Ma97], the road boudaries are modeled using a pair of tightly coupled parabolic curves. One novelty of our approach is to transform these constraints in polar coordinates which allows us to estimate the road edge parameters directly from recorded data. Unfortunately, maximum likelihood estimation yields estimates which are quite dependent on the off-road scatters. Another novelty of ourmethod is to construct an estimator which minimizes a cost criterion which is largely insensitive to the off-road scatter components. This criterion involves only the road pixels for which faithfull information is available, while its evaluation needs a reduced amount of calculations.
Our estimation procedure depends on a small number of parameters for which pertinent contraints can be established. This allows us to minimize our criterion by the means of pseudo-exhaustive search.
Numerical results are given for real data (L band radar images of a Southeast Michigan roadway) which indicate the accuracy and robustness of our approach. Several directions, concerning both the improvement of the numerical complexity and the robustification of the method, are actually under study.
Mila Nikolova received the Ph.D. degree in Signal Processing from the Universite de Paris-Sud in 1995. Currently, she is teaching and research assistant at the Universite Paris V. Her research interests are in inverse problems and image reconstruction.