Abstract
Lane detection technology is a critical component of autonomous driving, enabling precise recognition of road markings to achieve accurate navigation and stable vehicle operation in complex traffic environments. However, existing lane detection algorithms encounter some problems such as the need for presetting of anchor box parameters, cumbersome preprocessing and post-processing of datasets. To deal with these challenges, we propose an innovative lane detection algorithm called GaussianLaneNet. Firstly, to address the limitations of the FCOS algorithm in computing the confidence of lane center points, we introduce the use of a Gaussian distribution to model the confidence map of lane center points. Subsequently, we integrate the obtained confidence map with a global semantic map to enhance detection accuracy. Finally, we propose an adaptive threshold method to enable the model to learn the threshold autonomously, aiming to reduce the number of candidate samples effectively. We conducted experiments on the CULane dataset, and the results showcase that our method achieves an impressive F1 score of 80.85%, surpassing the performance of the majority of existing lane detection models.