Drivable Area and Lane Line Detection Model based on Semantic Segmentation

Aiming to address the problems of inadequate real-time performance and imprecise lane line and drivable area detection for autonomous vehicles, a semantic segmentation-based drivable areas and lane line detection model is proposed. This model consists of a Backbone, Neck, Drivable Area Head, and Lane Line Head. The Backbone extracts information from the input image, while the Neck combines features of various scales using the FPN structure to enhance feature extraction capability. The Drivable Area Head and Lane Line Head transform the width and height of the feature map back to the resolution of the input image. The CBS module is utilized to transform the channel number into two channels corresponding to the number of categories, and the Sigmoid layer generates the semantic probability output of the segmentation of the lane lines and drivable area. Finally, experiments are conducted using the BDD100K dataset, and results demonstrate that the drivable area and lane line detection model proposed in this work has good detection ability.