Edge-Based Detection of Structured Road Boundaries for Autonomous Land Vehicle

Abstract

In this paper, a vision scheme with feature-oriented image processing is proposed for structured road boundaries detection. Two criteria are adopted to select the candidates of road boundary edges. The first criterion is the interested range of edge direction and the other one is the minimum size of connected road edge pixels for the local continuity of road boundary. The noises, which can not satisfy the two criteria simultaneously, will be eliminated after performing the proposed image processing. For detecting a segment of anticipated road boundary, our vision scheme decides an area-of-interest subimage and a set of goal-driven parameters as the inputs of the feature-oriented image processing. The extracted boundary in the image is them represented by a straight line segment and used to update the road model. Experimental results show that this vision scheme is successful in detecting the road boundaries under various road environments, even with heavy interference, shadows, and changing illumination due to weather conditions.