Lane detection by estimating and using restricted search space in Hough domain

Abstract

Lane detection is one of the key elements for driving assistance systems. In this paper, a robust lane detection algorithm using restricted search space in Hough domain is presented. Although, it is not necessary to restrict the search space, many algorithms apply some restriction on some parameters (i.e. expected Vanishing point (Vp) position and the maximum deviation on Vp in both horizontal and vertical direction, expected road width (in pixels) and the maximum deviation on its width) to eliminate some of the unlikely detection results and prune some of the misdetections for more accurate results. However, conventionally, these parameters are manually selected and given to the algorithm as an input and same parameters cannot be used for different experimental set-up. In this algorithm, an alternative approach is proposed to automate this process. Novelty on this paper is in two folds. First, the proposed algorithm uses a set of road images for a given experimental set-up and statistically estimates both expected position and the expected deviation on Vp, and the road width. Second, a mask is created in Hough domain to efficiently apply necessary restrictions to both Road width and the Vp position. While this approach is applicable to any video sequence taken by any experimental set-up, it also benefits from restricting search space for more accurate detection results.The proposed algorithm is tested for 1000 highway road images and for only 28 road markings are miss detected. Revealing detection rate of more than 97%.