Height restriction barriers detection from traffic scenarios using stereo-vision

Abstract

Height restriction barriers detection is important usually for trucks' driving assistance systems. In this paper we propose a novel approach that uses stereo-vision and combines intensity and depth information for height barriers detection and tracking. High quality stereo-reconstruction is carried out by the SORT-SGM algorithm. Canny edges are extracted from 2D intensity image and filtered out by the 3D information. Horizontal lines are determined using the Hough transformation on the filtered edges and then validated by an intensity correlation approach taking into consideration that usually height restriction barriers have a repetitive textural pattern. Neighboring lines are clustered together in order to form the barrier region of interest. The height of the barrier is also approximated from the 3D points that belong to the barrier's region of interest. SURF features are extracted for the detected barriers from the intensity image and used for tracking them across frames. The whole height restriction barriers detection system performs real time with high accuracy results.