<p>Lane detection is a crucial component of visual navigation systems in intelligent vehicles. Current lane detection technology faces the dual challenges of real-time performance and accuracy, particularly in everyday scenarios where lane markings are often partially or completely occluded by objects, and thus significantly limits the development and advancement of autonomous driving technologies. In practice, lane lines inherently exhibit a degree of left–right symmetry, which would be helpful to improve the lane detection performance and accuracy. Specifically, by exploiting lane-line symmetry we can recover an occluded lane by horizontally flipping the visible lane and fusing the result. However, the symmetry of lane lines is not intuitive in most auto-driving scenarios. Addressing the challenges of real-time performance and accuracy, particularly in scenarios with occluded lane markings, we propose ABLaneNet—an adaptive lane detection method utilizing Bézier curve modeling. By exploiting lane-line symmetry and introducing an adaptive flipping fusion module, we mitigate feature bias caused by asymmetry. Furthermore, our non-uniform sampling loss method leverages the inherent properties of Bézier curves. Experiments on CULane, TuSimple and LLAMAS datasets demonstrate ABLaneNet’s superior performance, achieving 95.81% accuracy at 241 FPS, providing a reliable technical guarantee for advanced driver assistance and autonomous driving systems. Codes of our model are available at: <a href="https://doi.org/10.5281/zenodo.17301092">https://doi.org/10.5281/zenodo.17301092</a></p>