def meshify(points, n_surfaces=None):
"""Returns vertex and normal coordinates for a 3D mesh model from an Nx3 array of points after filtering.
Args:
-points (Nx3 Numpy Array): Data to be fit to a model.
-n_surfaces: If none, many different models with different numbers of surfaces will be compared.
Returns:
-vertices
-normals
"""
# Remove Obviously Bad Points according to how far away from main cluster they are
histmask = np.ones(points.shape[0], dtype=bool) # Initializing mask with all True values
for coord in range(3):
histmask &= filters.hist_mask(points[:, coord], keep='middle')
points_f = points[histmask, :]
# Get the normals of the N-Neighborhood around each point, and filter out points with lowish planarity
normals_f, explained_variances = normal_nearest_neighbors(points_f)
# Histogram filter: take the 70% best-planar data to model.
normfilter = filters.hist_mask(explained_variances[:, 2], threshold=.7, keep='lower')
points_ff = points_f[normfilter, :]
normals_ff = normals_f[normfilter, :]
ceiling_height = points_ff[:, 1].max() + .005
# Fit the filtered normal data using a gaussian classifier.
min_clusters = n_surfaces if n_surfaces else 4
max_clusters = n_surfaces + 1 if n_surfaces else 15
model = cluster_normals(normals_ff, min_clusters=min_clusters, max_clusters=max_clusters)
# Get normals from model means
surface_normals = model.means_ # n_components x 3 normals array, giving mean normal for each surface.
# Calculate mean offset of vertices for each wall
ids = model.predict(normals_ff) # index for each point, giving the wall id number (0:n_components)
surface_offsets = np.zeros_like(surface_normals)
for idx in range(len(surface_normals)):
surface_offsets[idx, :] = np.mean(points_ff[ids==idx, :], axis=0)
assert not np.isnan(surface_offsets.sum()), "Incorrect model: No Points found to assign to at least one wall for intersection calculation."
## CALCULATE PLANE INTERSECTIONS TO GET VERTICES ##
vertices, normals = get_vertices_at_intersections(surface_normals, surface_offsets, ceiling_height)
return vertices, normals
# If the experimenter needs to enter the room, give them a bit of time to get inside.
if args.human_scan:
time.sleep(5)
# Collect random points for calibration.
motive.load_project(args.motive_projectfile)
hardware.motive_camera_vislight_configure()
screenPos, pointPos = random_scan(window, scene, n_points=args.n_points)
print("Size of Point Data: {}".format(pointPos.shape))
# Remove Obviously Bad Points according to how far away from main cluster they are
histmask = np.ones(pointPos.shape[0], dtype=bool) # Initializing mask with all True values
for coord in range(3):
histmask &= filters.hist_mask(pointPos[:, coord], keep='middle')
pointPos = pointPos[histmask, :]
screenPos = screenPos[histmask, :]
print("Size of Point Data after hist_mask: {}".format(pointPos.shape))
# Project a few points that the experimenter can make rays from (by moving a piece of paper up and down along them.
# Don't include human data, because its non-gaussian distribution can screw things up a bit.
# TODO: Figure out how to properly get human-scanned projector calibration data in so OpenCV gets better estimate.
if args.human_scan:
ray_scan(window)
# Close Window
window.close()
# Early Tests