Abstract: With the rapid development of autonomous driving technology, lane detection, as a core function of advanced driver assistance systems (ADAS), has a direct impact on driving safety through its accuracy and robustness. To address the limitations of traditional methods that rely on manual feature extraction and the insufficient performance of existing deep learning models in complex scenarios, this paper proposes an improved dual-branch lane detection method based on ResNet-18. The method transforms lane detection into an instance segmentation task by designing a collaborative operation between the auxiliary branch and the main branch, while introducing a compound loss function to optimize model performance. Experimental results demonstrate that the improved model exhibits superior adaptability in complex environments, effectively addressing challenges such as vehicle occlusion and illumination variations. This work provides a novel solution for lane detection in challenging real-world scenarios.

Key words: lane detection; ResNet; dual-branch network; instance segmentation

摘要： 随着自动驾驶技术的快速发展，车道线检测作为高级驾驶辅助系统（ADAS）的核心功 能，其准确性与鲁棒性直接影响行车安全。针对传统方法依赖人工特征、现有深度学习模型 在复杂场景下性能不足的问题，文章提出一种基于 ResNet-18 改进的双分支网络车道线检测 方法，通过设计辅助分支与主干分支协同工作，将车道线检测任务转化为实例分割问题，并 引入复合损失函数优化模型性能。实验结果表明，改进后的模型在复杂场景下具有更好的适 应性，能够有效应对车辆遮挡和光照变化等挑战，为复杂环境下的车道线检测提供了新的解 决方案。

关键词: 车道线检测；ResNet；双分支网络；实例分割