curb_aggregate requires a point cloud topic. It combines a few clouds together (using tf lookup) which makes the cloud denser and hence easier to process. It needs a config file (example in config folder)
There are two methods
- using normal and curvature threshold
- dividing a surrounding point cloud to an image like collection of cells, using height difference between min and max z in each cell to determine if the point in a cell belongs to the curb
Both methods are still quite noisy