Improving Lane Annotation in Autonomous Driving Data Sets with Classical Computer Vision Techniques

Abstract

Autonomous driving systems rely on accurate and reliable lane detection to safely navigate roads. In this paper, we propose a method for improving lane annotation in autonomous driving data sets using classical computer vision techniques. The proposed method combines the Hough transform and linear curve fitting to detect and smooth the positions of lane markings in a video stream. We evaluate the performance of the proposed method on the Berkeley DeepDrive (BDD) dataset and compare it to the ground truth annotations. Results show that the proposed method achieves a high level of accuracy and robustness in lane detection, and can effectively improve lane annotation in autonomous driving data sets. Our method provides a valuable tool for training and evaluating autonomous driving systems, and can also be applied to improve annotation in different datasets.