Lane detection is a critical component of autonomous driving technologies that face challenges such as varied road conditions and diverse lane orientations. In this study, we aim to address these challenges by proposing PolyLaneDet, a novel lane detection model that utilizes a freeform polyline, termed ‘polylane,’ which adapts to both vertical and horizontal lane orientations without the need for post-processing. Our method builds on the YOLOv4 architecture to avoid restricting the number of detectable lanes. This model can regress both vertical and horizontal coordinates, thereby improving the adaptability and accuracy of lane detection in various scenarios. We conducted extensive experiments using the CULane benchmark and a custom dataset to validate the effectiveness of the proposed approach. The results demonstrate that PolyLaneDet achieves a competitive performance, particularly in detecting horizontal lane markings and stop lines, which are often omitted in traditional models. In conclusion, PolyLaneDet advances lane detection technology by combining flexible lane representation with robust detection capabilities, making it suitable for real-world applications with diverse road geometries.

Keywords: Lane detection, CNN, Deep learning

The authors declare no conflicts of interest regarding this study.