Lane line detection method based on semantic segmentation

The invention belongs to the technical field of lane line detection, and relates to a semantic segmentation-based lane line detection method, which comprises the following steps of: acquiring an image, acquiring a lane line picture shot by an automobile data recorder, designing a lane feature extraction network which is divided into two parts, namely a main branch and an auxiliary branch, after the image is input, selecting an ROI (Region of Interest) to perform preprocessing operation, the main branch structure performs feature extraction and feature overturning fusion operation on the image, and finally outputs a lane line detection result; the auxiliary branch performs cascading operation on the images subjected to feature extraction in the main branch to obtain a lane line segmentation result; wherein the auxiliary branch is used for assisting detection of lane lines, only participates in a training process and does not participate in a reasoning process; then designing energy loss to carry out joint trai