Vanishing points estimation and lanes identification based on Bayesian posterior probability

This paper presents a novel scheme for Vanishing Points (VPs) estimation and lanes identification through monocular images of mobile robots. VPs detection based on probable vanishing direction hypothesizes and Bayesian posterior probability verification in image Hough space is a foremost contribution. VPs estimation is an optimal resolution based on a weighted objective function. The selected linear samples supervised by estimated VPs are clustered based on the gradient direction of features to separate different lanes. Finally the lanes are identified through the identification function. Especially, the branch road with many vanishing points can be identified. Our scheme is tested on real datasets collected from an intelligent vehicle. Experimental results demonstrate VPs and lanes can be detected accurately in the challenging structured and semi-structured complex road scenarios.