better smoothing algorithm

1 files changed, 84 insertions, 0 deletions

diff --git a/lib/utils/smoothing.py b/lib/utils/smoothing.py
new file mode 100644
index 00000000..9d43a9f1
--- /dev/null
+++ b/lib/utils/smoothing.py
@@ -0,0 +1,84 @@
+import numpy as np
+from scipy.interpolate import splprep, splev
+
+from .geometry import Point, coordinate_list_to_point_list
+from ..stitches.running_stitch import running_stitch
+from ..debug import debug
+
+
+def _remove_duplicate_coordinates(coords_array):
+    """Remove consecutive duplicate points from an array.
+
+    Arguments:
+        coords_array -- numpy.array
+
+    Returns:
+        a numpy.array of coordinates, minus consecutive duplicates
+    """
+
+    differences = np.diff(coords_array, axis=0)
+    zero_differences = np.isclose(differences, 0)
+    keepers = np.r_[True, np.any(zero_differences == False, axis=1)]  # noqa: E712
+
+    return coords_array[keepers]
+
+
+@debug.time
+def smooth_path(path, smoothness=1.0):
+    """Smooth a path of coordinates.
+
+    Arguments:
+        path -- an iterable of coordinate tuples or Points
+        smoothness -- float, how much smoothing to apply.  Bigger numbers
+            smooth more.
+
+    Returns:
+        A list of Points.
+    """
+    from ..debug import debug
+
+    if smoothness == 0:
+        # s of exactly zero seems to indicate a default level of smoothing
+        # in splprep, so we'll just exit instead.
+        return path
+
+    # Smoothing seems to look nicer if the line segments in the path are mostly
+    # similar in length.  If we have some especially long segments, then the
+    # smoothed path sometimes diverges more from the original path as the
+    # spline curve struggles to fit the path.  This can be especially bad at
+    # the start and end.
+    #
+    # Fortunately, we can convert the path to segments that are mostly the same
+    # length by using the running stitch algorithm.
+    path = running_stitch(coordinate_list_to_point_list(path), 5 * smoothness, smoothness / 2)
+
+    # splprep blows up on duplicated consecutive points with "Invalid inputs"
+    coords = _remove_duplicate_coordinates(np.array(path))
+    num_points = len(coords)
+
+    if num_points <= 3:
+        # splprep throws an error unless num_points > k
+        return path
+
+    # s is explained in this issue: https://github.com/scipy/scipy/issues/11916
+    # the smoothness parameter limits how much the smoothed path can deviate
+    # from the original path.  The standard deviation of the distance between
+    # the smoothed path and the original path is equal to the smoothness.
+    # In practical terms, if smoothness is 1mm, then the smoothed path can be
+    # up to 1mm away from the original path.
+    s = num_points * (smoothness ** 2)
+
+    # .T transposes the array (for some reason splprep expects
+    # [[x1, x2, ...], [y1, y2, ...]]
+    tck, fp, ier, msg = splprep(coords.T, s=s, k=3, nest=-1, full_output=1)
+    if ier > 0:
+        debug.log(f"error {ier} smoothing path: {msg}")
+        return path
+
+    # Evaluate the spline curve at many points along its length to produce the
+    # smoothed point list.  2 * num_points seems to be a good number, but it
+    # does produce a lot of points.
+    smoothed_x_values, smoothed_y_values = splev(np.linspace(0, 1, int(num_points * 2)), tck[0])
+    coords = np.array([smoothed_x_values, smoothed_y_values]).T
+
+    return [Point(x, y) for x, y in coords]