Lane Detection and Tracking based on Best Pairs of Lane Markings: Method and Evaluation

This study proposes a novel lane detection and tracking algorithm. Firstly, feature extraction is performed by convolving a grayscale bird's eye view image with a 1 dimensional top-hat kernel and applying a histogram computation afterwards. Next, the Hough Transform is applied to detect lines that are then merged based on their geometrical characteristics. Then, the best lane pair is selected based on a custom cost function. Lastly, polynomial lane model parameters are estimated for the best lane pair and tracked by a Kalman Filter. Our computational results show that the proposed method can detect lanes in complex cases and increase the lane detection accuracy.