Intelligent lane detection and path prediction for autonomous driving in varied weather

Abstract

Rapidly advancing intelligent and autonomous driving systems demand reliable computer vision-based perception technology, particularly for safe path detection in various weather and road conditions, essential for efficient vehicle navigation. This paper proposes a novel lane detection technique utilizing a pre-trained Keras-based CNN model capable of identifying the path ahead under challenging lighting and weather situations, such as nighttime and heavy rain, using videos acquired with a monocular camera. Furthermore, we address the issue of lane detection when lines are illuminated by vehicle headlights or streetlights, even under severely reduced visibility conditions caused by heavy rainfall, using a color filtering technique. We propose a path projection technique that integrates the widely used slope calculation method with convolutional neural networks (CNN). The Keras model facilitates the detection of lane lines, enabling the calculation of the center trajectory based on the identified road lanes. The projection technique demonstrates effective performance in low visibility and adverse weather conditions. The experimental results show that the presented algorithms effectively detect road lanes and predict paths in multiple weather conditions.