Journal Search Engine
The Journal of The Korea Institute of Intelligent Transport Systems Vol.24 No.5 pp.200-212
카메라 시야각 변화에 강건한 차선 인식 방법
Robust Lane Detection Method against Viewpoint Variations
Abstract
Fast and accurate lane detection is a crucial task for the safe operation of autonomous vehicles. However, differences in vehicle height and camera installation angles cause variations in the vertical viewpoint of input images, leading to a domain shift that degrades the accuracy of lane detection networks. This paper proposes a novel lane detection method that enhances robustness to viewpoint variations by integrating an offset head and alignment block into existing networks. The proposed approach was designed in a block-based manner, allowing it easy integration into pre-trained network without requiring full retraining, while effectively mitigating the impact of a domain shift under diverse camera viewpoints. Evaluations on the custom expanded testset constructed from the CULane benchmark with diverse view conditions showed stable precision, achieving an F1 Score of 69.57, which is approximately 2.8 times higher than the baseline score of 25.19, even when the portion of the image occupying the lane is reduced by 50%. These results highlight the practicality of the proposed method in real-world deployment environments with variable viewpoints.
