Road lanes detection in challenging conditions

Abstract

This thesis is the final report of the graduation project conducted as part of the Master 2 program in Artificial Intelligence for the Internet of Things (AI4IoT). The project focuses on the development of a classical lane detection system for autonomous driving applications, with the main objective of identifying lane lines in road scenes using traditional computer vision techniques. The system relies on a sequence of image processing steps, including grayscale conversion, edge detection, region masking, and Hough Transform, to detect lane boundaries in each video frame. The TuSimple dataset was used to evaluate the method's effectiveness in real-world highway scenarios. A basic averaging technique was also applied to improve temporal stability across consecutive frames. This project combines knowledge in computer vision, image processing, and real-time system design. It demonstrates how classical techniques, despite the rise of deep learning, can still offer efficient and interpretable solutions for specific tasks such as lane detection in structured driving environments. The work contributes to the broader field of intelligent transportation systems by providing a lightweight and explainable alternative to data-driven approaches.