Reliable lane detection under adverse weather and illumination remains a critical challenge for low-cost autonomous driving systems. This study proposes a Binary Line Segment Filter (BLSF) that exploits lane geometry for robust and efficient feature extraction. Unlike traditional intensity-based methods, BLSF maintains high accuracy across challenging conditions—including high curvature, strong backlighting, nighttime low contrast, and heavy rain—achieving an average detection rate of 95%. Remarkably, performance under strong backlight improves from 3% to 93%. The algorithm operates at 32.7 ms per frame on a 2 GHz ARM processor, fulfilling real-time constraints without relying on deep learning.
Tsai et al. (Mon,) studied this question.