When a line has several branches, we still need a single linestring representing the line geom. lowess regression is used when the line has branches, but we need something better. Lowess is too coarse. Maybe routing the lowess line using projected points over that line and routing from one to the other through a osmnx line.
We could also use graphs and get the common part and the branching, using the common part geom as it is and producing a lowess with routing over the branching, always going back to the same node that branches.
When a line has several branches, we still need a single linestring representing the line geom. lowess regression is used when the line has branches, but we need something better. Lowess is too coarse. Maybe routing the lowess line using projected points over that line and routing from one to the other through a
osmnxline.We could also use graphs and get the common part and the branching, using the common part geom as it is and producing a lowess with routing over the branching, always going back to the same node that branches.