Efficient vanishing point detection method in complex urban road environments

Abstract

Detecting the vanishing point in a road image is important for robot navigation, intelligent transportation and other fields. This study proposes a new vanishing point detection method that uses the vertical information in complex urban road and street environments. First, all straight lines in the road image are detected based on curvature scale space and principal component analysis methods. Second, a new road region extraction method is proposed to support vanishing point detection. In this method, the image is decomposed into approximate sky, vertical and road regions using the estimation envelopes of the vertical lines. Third, the straight lines in the road region are extracted using a path perspective triangle and line length limits. The straight lines are then categorised into groups using proposed grouping strategies. Finally, vanishing point candidates are calculated from paired lines extracted from different groups and from the same group, and the final vanishing point is then estimated for the urban road scene using the mean shift clustering method. The experimental results show that the proposed algorithm can estimate the vanishing point accurately and efficiently in complex urban road environments, despite interference from vehicles and pedestrians and on curved and unstructured roads.