A Real-Time, Training-Free Lane Detection Approach Robust to Day–Night Illumination Changes On CPU-Only Platforms

Lane-line detection Night-time driving Advanced driver-assistance systems Region-of-interest masking Autonomous driving

Authors

  • Marwa K. Hasan Technical Engineering College - Kirkuk, Northern Technical University, Kirkuk, 36001, Iraq
  • Lana O. Alhamawndi Technical Engineering College - Kirkuk, Northern Technical University, Kirkuk, 36001, Iraq
  • Alaan Ghazi Department of Cybersecurity Techniques Engineering, College of Computer Engineering Technology and Artificial Intelligence - Kirkuk, Northern Technical University, Mosul, 41001, Iraq
  • Lamees S. Ahmed Technical Engineering College - Kirkuk, Northern Technical University, Kirkuk, 36001, Iraq
  • Ali J. Mohammed College of Oil and Gas Technology Engineering / Kirkuk, Northern Technical University, Kirkuk, Iraq
March 20, 2026
March 24, 2026

Downloads

This paper presents a lightweight, training-free lane-line detection pipeline for daytime and night-time driving videos implemented in Python using OpenCV. The method performs illumination-aware pre-processing using CLAHE, and applies mild gamma correction in LAB space for low-light scenes. A trapezoidal road region-of-interest is then masked to suppress background clutter, followed by adaptive edge extraction using Canny with data-driven thresholds. Candidate lane segments are obtained via the probabilistic Hough transform and separated into left/right groups using slope-based filtering. The final lane boundaries are estimated by length-weighted averaging of line parameters and rendered as an overlay to visualize the drivable lane region and heading direction. Experiments on day and night recordings captured on the same road segments demonstrate stable lane detection under shadows, headlight bloom, and worn markings. The proposed approach is suitable for CPU-only embedded/mobile deployment and can support lane-departure warning and lane-keeping assistance in ADAS and autonomous driving applications.