Lane detection is essential for automated driving and advanced driver assistance systems (ADAS), including functions like lane-keeping, lane departure warning, and forward collision warning. Nevertheless, adverse weather conditions such as heavy rain or snow can substantially affect the effectiveness of lane detection systems, posing challenges in real-world scenarios. To address these challenges, we proposed an image augmentation strategy for the Cross-Layer Refinement Network-based lane detection system in adverse weather conditions. The benefits of the augmentation strategy are evaluated by training and testing the models with and without augmentation. The proposed method outperforms the non-augmented model on well-established benchmark datasets such as CULane and TuSimple. As shown in the experimental results, 12.1% enhancement in average F1@50 accuracy and 6.3% improvement in F1 accuracy are achieved. This confirms the effectiveness of the augmented cross-layer refinement network in achieving robust lane detection performance under various adverse weather conditions.
Augmented Cross Layer Refinement Network-Based Lane Detection in Adverse Weather Conditions
24.06.2024
3784330 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Vision-based Lane Recognition under Adverse Weather Conditions Using Optical Flow
British Library Conference Proceedings | 2003
|