Real-time detection of lanes and boundaries by autonomous vehicles

A method comprising:

• receiving image data representative of an image of an environment;
• and
• generating a polygon associated with the image by:
• identifying first vertices that correspond to a boundary within the environment;
• generating first polylines extending between adjacent first vertices;
• based at least in part on the first polylines, generating second vertices adjacent the first vertices, the second vertices being spaced from the first vertices along a direction perpendicular to the first polylines;
• generating a second polyline extending between the second vertices;
• generating third polylines extending between corresponding first vertices and second vertices;
• and
• training a neural network using the image data, the first polylines, the second polyline, and the third polylines as ground truth data.

A method comprising:

• receiving image data representative of an image of a driving surface;
• receiving annotations corresponding to locations of at least one of lane markings or boundaries of the driving surface;
• applying one or more first transformations to the image to generate a transformed image;
• applying one or more second transformations corresponding to the one or more first transformations to each of the annotations to generate transformed annotations;
• and
• using the image, the annotations, the transformed image, and the transformed annotations as ground truth data to train a neural network to identify pixels within images that correspond to at least one of the lane markings or the boundaries.