Near real-Time ego-lane detection in highway and urban streets

In this paper, we present a near real time approach to lane detection in highway and urban streets using images captured from monocular cameras as an input. The proposed method includes four main steps. First, the input image is transformed to a bird's eye view image by using warp perspective mapping. Second, an edge image is produced by applying a custom edge detection and morphology operation on the bird's eye view image. Then, initial lines are extracted by using a Hough transform and a sliding window technique. Finally, the ego-lanes are detected by performing a custom lane fitting method on the initial lines. The experimental results showed that our algorithm can detect the ego-lanes in both ordinary and some complex situations such as when symbols and Hangul markings appear on the road and when pedestrians walk in front of the car. Using image sequences provided by Hyundai Motor Company, our proposed method provided a relatively accurate result while performing at a rate of 20 fps on large image (1208 × 672).