Performing semantic segmentation of 3d data using deep learning

A method, comprising:

• subdividing a three-dimensional point cloud into a plurality of overlapping columnar subdivisions and assigning semantic labels to points of each subdivision using a trained model, wherein said semantic labels comprise one or more labels corresponding to a roof structure;
• removing, based on the semantic labels, points that are not classified as a roof structure from the subdivisions to generate filtered subdivisions;
• and
• determining roof geometry of an outdoor structure based on the filtered subdivisions.

A system, comprising:

• a processor configured to:
• subdivide a three-dimensional point cloud into a plurality of overlapping columnar subdivisions and assign semantic labels to points of each subdivision using a trained model, wherein said semantic labels comprise one or more labels corresponding to a roof structure;
• remove, based on the semantic labels, points that are not classified as a roof structure from the subdivisions to generate filtered subdivisions;
• and
• determine roof geometry of an outdoor structure based on the filtered subdivisions;
• and
• a memory coupled to the processor and configured to provide the processor with instructions.

A computer program product embodied in a non-transitory computer readable medium and comprising computer instructions for:

• subdividing a three-dimensional point cloud into a plurality of overlapping columnar subdivisions and assigning semantic labels to points of each subdivision using a trained model;
• removing, based on the semantic labels, points classified as vegetation from the subdivisions to generate filtered subdivisions;
• and
• determining roof geometry of an outdoor structure based on the filtered subdivisions.