Autonomous Vehicle Lane Detection: A Deep Learning-Based Approach for Enhanced Road Safety

Abstract

Lane detection is a fundamental task in autonomous vehicle systems, enabling precise lane-keeping and safe navigation. This research introduces a comprehensive lane detection framework that integrates computer vision techniques with real-time processing for improved accuracy and efficiency. The proposed system begins with camera calibration to correct lens distortions, ensuring that the captured frames represent accurate spatial information. Adaptive thresholding techniques are then applied to highlight lane features, filtering out irrelevant details and reducing computational complexity. Perspective transformation is used to generate a bird’s-eye view of the road, allowing for better interpretation of lane geometry. To identify lane boundaries effectively, Canny edge detection extracts prominent edges, while the Hough transform identifies straight lines that correspond to lane markers. The system is implemented using OpenCV and tested in a cloud-based environment, facilitating real-time performance analysis and ensuring scalability for different datasets and video inputs. Experimental evaluations demonstrate that the system consistently achieves high accuracy in identifying lane boundaries under diverse road and lighting conditions. Additionally, the framework maintains computational efficiency, making it a practical solution for integration into real-time autonomous driving applications. This research contributes to the advancement of lane detection technology by offering a reliable, scalable, and efficient approach that can be adapted to various real-world scenarios.