import cv2
import numpy as np
import os
[docs]
def create_warpfile(projection_height: int, projection_width: int, projector_dir: str):
"""
Function to create and save a warp mesh file
Parameters
----------
projection_height : int
Height of the projection in pixels (eg. 1080).
projection_width : int
Width of the projection in pixels (eg. 1920).
projector_dir : str
Directory where the roi file is stored and
where the warp file will be stored
Returns
-------
warpfile : str
Absolute path to the created warp file.
The name of the warpfile will be
"warpmesh_{projection_height}x{projection_width}.data"
and will be stored under projector_dir
"""
# get the rois
roi_filename = "rois_{}x{}.npy".format(projection_height, projection_width)
if not os.path.exists(os.path.join(projector_dir, roi_filename)):
raise FileNotFoundError("ROI file not found: %s" % os.path.join(projector_dir, roi_filename))
rois = np.load(os.path.join(projector_dir, roi_filename))
# scale the rois such that bottom left is [-1, -1], top right is [1, 1]
rois_scaled = np.array(rois, dtype=float)
rois_scaled[:, :, 0] = (rois[:, :, 0] - (projection_width / 2)) / (projection_width / 2)
rois_scaled[:, :, 1] = ((projection_height / 2) - rois[:, :, 1]) / (projection_height / 2)
nb_screens = rois.shape[0]
frame_height, frame_width = 1 / nb_screens, 1
# get the transformation matrices
transformation_matrices = get_transformation_matrices(
image_height=frame_height, image_width=frame_width, proj_rois=rois_scaled
)
# create a warping mesh
# http://paulbourke.net/dome/warpingfisheye/
# warped image coordinates: (x,y) varies from [-1,-1] (bottom left) to [1,1] (top right)
# Note: in Paul Bourke's website, warped coordinates varied from
# [-aspect_ratio, -1] (bottom left) to [aspect_ratio,1] (top right),
# but for psychpy [-1,-1] to [1,1] works
# original image coordinates: (u,v) varies from [0,0] (bottom left) to [1,1] (top right)
# intensities i varies from 0 to 1
node_x = np.linspace(-1, 1, projection_width, dtype="float32")
node_y = np.linspace(-1, 1, projection_height, dtype="float32")
node_grid_x, node_grid_y = np.meshgrid(node_x, node_y)
node_positions_xy = np.concatenate((node_grid_x.reshape(-1, 1), node_grid_y.reshape(-1, 1)), axis=1)
mesh_coordinates_uv = np.zeros(node_positions_xy.shape, dtype="float32")
intensities = np.zeros((node_positions_xy.shape[0], 1), dtype="float32")
projection_image = np.zeros((projection_height, projection_width), dtype="uint8")
for iscreen in range(rois_scaled.shape[0]):
projection_mask = projection_image.copy()
# get the pixel ids in the roi
roi_mask = cv2.fillConvexPoly(projection_mask, rois[iscreen], 1)
roi_node_idx = np.vstack(np.where(roi_mask)).T
# flip y
roi_node_idx[:, 0] = projection_height - roi_node_idx[:, 0]
roi_node_idx_flat = roi_node_idx[:, 0] * projection_width + roi_node_idx[:, 1]
# get node positions in roi
roi_node_xy = node_positions_xy[roi_node_idx_flat]
# convert from euclidean to homogeneous coordinates
roi_node_xy_h = np.concatenate((roi_node_xy, np.ones((roi_node_xy.shape[0], 1))), axis=1)
# apply transformation to mesh coordinates
roi_mesh_uv_h = roi_node_xy_h @ transformation_matrices[iscreen].T
# convert back to euclidean coordinates
roi_mesh_uv = roi_mesh_uv_h[:, [0, 1]] / roi_mesh_uv_h[:, [2]]
mesh_coordinates_uv[roi_node_idx_flat] = roi_mesh_uv
intensities[roi_node_idx_flat] = 1
# xyuvi coordinates
warp_mesh_coordinates = np.concatenate((node_positions_xy, mesh_coordinates_uv, intensities), axis=1)
# save to file
warpfile = os.path.join(projector_dir, "warpmesh_{}x{}.data".format(projection_height, projection_width))
filetype = 2
with open(os.path.join(projector_dir, warpfile), "w+") as f:
f.write("%d\n" % (filetype))
f.write("%d %d\n" % (projection_width, projection_height))
np.savetxt(f, warp_mesh_coordinates, fmt="%0.5f")
return warpfile
etho