Determination of lane position is an important component in the driver-assistance systems to provide meaningful and
consistent road shape information for navigation purpose. Extraction of lane markings is a key component to detection of
lane position. In this paper, we propose a new algorithm for extraction of lane markings based on steerable filters, which
have been used to analyze local orientation patterns in imagery. Steerable filters are efficient for extraction of lane
markings because by computing only three separable convolutions, we can extract a wide variety of lane markings. The
steerable filters used for detection of lane are based on second derivatives of two-dimensional Gaussians. Such filters
based on even-order derivatives are symmetric. While symmetry produces orientation responses that are periodic with
period , independent of image structure, we present a more general asymmetric steerable function that alleviates this
problem. The algorithm is able to provide robust and accurate extraction of lane markings under varying lighting and
road conditions.
|