feat: draft fvt and ilv

This commit is contained in:
gdamms 2023-01-11 13:18:31 +01:00
parent 322d336ce8
commit 06dcf8fb4f
5 changed files with 286 additions and 14 deletions

2
.gitignore vendored Normal file
View file

@ -0,0 +1,2 @@
__pycache__
data

141
fvi.py Normal file
View file

@ -0,0 +1,141 @@
import numpy as np
def fast_voxel_intersect(start, end, origin, step) -> tuple[list, list]:
"""Compute the voxels intersected by a line segment.
Args:
start (array-like): start point of line segment
end (array-like): end point of line segment
origin (array-like): origin of voxel grid
step (array-like): step size of voxel grid
Returns:
list: list of intersection points
list: list of intersected voxels
"""
# Convert to numpy arrays
start = np.asarray(start)
end = np.asarray(end)
origin = np.asarray(origin)
step = np.asarray(step)
# Translate line segment to voxel grid
start = start - origin
end = end - origin
# Initialize list of intersected voxels
intersections = []
voxels = []
# Compute direction of line segment
direction = end - start
global_distance = np.linalg.norm(direction, axis=0)
if global_distance == 0:
return intersections
direction = direction / global_distance
# Compute the sign of the direction
direction_signs = np.sign(direction)
is_positive = direction_signs > 0
is_negative = direction_signs < 0
# Initialize current position to start
position = start.copy()
# Main loop
while True:
# Compute the distance to the next boundaries
next_boundaries = np.divide(position + step * direction_signs, step)
distances = (is_positive * np.floor(next_boundaries) +
is_negative * np.ceil(next_boundaries)) * step - position
# Determine the nearest boundary to be reached
boundary_distances = np.abs(distances / direction)
clothest_boundary = np.argmin(boundary_distances)
clothest_boundary_distance = boundary_distances[clothest_boundary]
# Check if we are done
distance_to_end = abs((end[0] - position[0]) / direction[0])
if clothest_boundary_distance > distance_to_end:
break
# Update position
position = position + clothest_boundary_distance * direction
# Correct position to be on boundary
position[clothest_boundary] = round(
position[clothest_boundary] / step[clothest_boundary]) * step[clothest_boundary]
# Get corresponding voxel
boundary_reached_negativly = np.zeros(start.shape, dtype=bool)
boundary_reached_negativly[clothest_boundary] = is_negative[clothest_boundary]
voxel = np.floor(position) - boundary_reached_negativly * step
# Add voxel to list
intersections.append(position + origin)
voxels.append(voxel + origin)
return intersections, voxels
if __name__ == '__main__':
import matplotlib.pyplot as plt
def update_figure():
positions, voxels = fast_voxel_intersect(start, end, origin, step)
plt.clf()
# Plot hitted voxels
for voxel in voxels:
plt.fill([voxel[0], voxel[0] + step[0], voxel[0] + step[0], voxel[0]],
[voxel[1], voxel[1], voxel[1] + step[1], voxel[1] + step[1]],
color='#e25', alpha=0.5)
# Plot line segment
plt.plot([start[0], end[0]], [start[1], end[1]], 'k-')
plt.plot(start[0], start[1], 'go')
plt.plot(end[0], end[1], 'ro')
# Plot intersection points
for pos in positions:
plt.plot(pos[0], pos[1], 'bo')
# Plot voxel grid
plt.axis('equal')
plt.xlim((-10, 10))
plt.ylim((-10, 10))
xmin, xmax = plt.xlim()
ymin, ymax = plt.ylim()
plt.xticks(np.arange(xmin + origin[0],
xmax + origin[0] + step[0], step[0]))
plt.yticks(np.arange(ymin + origin[1],
ymax + origin[1] + step[1], step[1]))
plt.grid()
plt.draw()
def onclick(event):
global start, end
# if event.button == 1:
# start = np.array([event.xdata, event.ydata])
# elif event.button == 3:
# end = np.array([event.xdata, event.ydata])
start = np.random.rand(2) * 10 - 5
end = np.random.rand(2) * 10 - 5
update_figure()
# Define voxel grid
origin = np.array([.1, -.3])
step = np.array([1.0, 1.0])
# Define segment
start = np.random.rand(2) * 10 - 5
end = np.random.rand(2) * 10 - 5
# Plot
fig = plt.figure()
fig.canvas.mpl_connect('button_press_event', onclick)
update_figure()
plt.show()

102
intersec_line_voxel.py Normal file
View file

@ -0,0 +1,102 @@
import numpy as np
import matplotlib.pyplot as plt
from itertools import product
def check_line_voxel(
px, py, pz,
dx, dy, dz,
vx, vy, vz,
c
):
"""Check if a line intersects a voxel."""
# Compute the intersection bounds
kx1 = (px - dx) / vx
ky1 = (py - dy) / vy
kz1 = (pz - dz) / vz
kx2 = (px - dx + c) / vx
ky2 = (py - dy + c) / vy
kz2 = (pz - dz + c) / vz
# Order the bounds
kxmin = np.min(np.concatenate([
kx1[:, np.newaxis],
kx2[:, np.newaxis]
], axis=1), axis=1)
kymin = np.min(np.concatenate([
ky1[:, np.newaxis],
ky2[:, np.newaxis]
], axis=1), axis=1)
kzmin = np.min(np.concatenate([
kz1[:, np.newaxis],
kz2[:, np.newaxis]
], axis=1), axis=1)
kxmax = np.max(np.concatenate([
kx1[:, np.newaxis],
kx2[:, np.newaxis]
], axis=1), axis=1)
kymax = np.max(np.concatenate([
ky1[:, np.newaxis],
ky2[:, np.newaxis]
], axis=1), axis=1)
kzmax = np.max(np.concatenate([
kz1[:, np.newaxis],
kz2[:, np.newaxis]
], axis=1), axis=1)
# Check if the bounds overlap
kmax = np.min(np.concatenate([
kxmax[:, np.newaxis],
kymax[:, np.newaxis],
kzmax[:, np.newaxis]
], axis=1), axis=1)
kmin = np.max(np.concatenate([
kxmin[:, np.newaxis],
kymin[:, np.newaxis],
kzmin[:, np.newaxis]
], axis=1), axis=1)
return kmin <= kmax
c = 1.0
points = np.array([[x, y, z] for x, y, z in product(
np.arange(-5.0, 4.0, c),
np.arange(-5.0, 4.0, c),
np.arange(-5.0, 4.0, c))
])
while True:
fig = plt.figure()
ax = plt.axes(projection='3d')
d = np.random.rand(3) * 1 - 0.5
v = np.random.rand(3) * 1 - 0.5
px, py, pz = points[:, 0], points[:, 1], points[:, 2]
dx, dy, dz = d
vx, vy, vz = v
bool_vect = check_line_voxel(px, py, pz, dx, dy, dz, vx, vy, vz, c)
# plot cube
for i, (px, py, pz) in enumerate(points):
if not bool_vect[i]:
continue
ax.plot([px, px+c], [py, py], [pz, pz], 'b')
ax.plot([px, px+c], [py, py], [pz+c, pz+c], 'b')
ax.plot([px, px+c], [py+c, py+c], [pz, pz], 'b')
ax.plot([px, px+c], [py+c, py+c], [pz+c, pz+c], 'b')
ax.plot([px, px], [py, py+c], [pz, pz], 'b')
ax.plot([px, px], [py, py+c], [pz+c, pz+c], 'b')
ax.plot([px+c, px+c], [py, py+c], [pz, pz], 'b')
ax.plot([px+c, px+c], [py, py+c], [pz+c, pz+c], 'b')
ax.plot([px, px], [py, py], [pz, pz+c], 'b')
ax.plot([px, px], [py+c, py+c], [pz, pz+c], 'b')
ax.plot([px+c, px+c], [py, py], [pz, pz+c], 'b')
ax.plot([px+c, px+c], [py+c, py+c], [pz, pz+c], 'b')
# plot line
ax.plot([dx, dx+vx], [dy, dy+vy], [dz, dz+vz], 'g')
plt.show()

55
main.py
View file

@ -1,30 +1,57 @@
import cv2 import cv2
import numpy as np import numpy as np
from itertools import product
from rich.progress import track
from matrices_reader import * from matrices_reader import *
VOXEL_SIZE = 1e-3 VOXEL_SIZE = 2e-2
X_MIN, X_MAX = -2.0, 2.0 X_MIN, X_MAX = -2.0, 2.0
Y_MIN, Y_MAX = -2.0, 2.0 Y_MIN, Y_MAX = -2.0, 2.0
Z_MIN, Z_MAX = -2.0, 2.0 Z_MIN, Z_MAX = -2.0, 2.0
# grid = [[[
# 1 for z in np.arange(Z_MIN, Z_MAX, VOXEL_SIZE)
# ] for y in np.arange(Y_MIN, Y_MAX, VOXEL_SIZE)
# ] for x in np.arange(X_MIN, X_MAX, VOXEL_SIZE)
# ]
projection_matrices = matrices_reader('data/torus/matrices.txt') projection_matrices = matrices_reader('data/torus/matrices.txt')
nb_frame = len(projection_matrices) nb_frame = len(projection_matrices)
point = np.array([1.0, 0.0, 0.0, 1.0])
points = np.array([[x, y, z, 1.0] for x, y, z in product(
np.arange(X_MIN, X_MAX, VOXEL_SIZE),
np.arange(Y_MIN, Y_MAX, VOXEL_SIZE),
np.arange(Z_MIN, Z_MAX, VOXEL_SIZE))])
background = np.array([18, 18, 18])
for k in range(nb_frame): for k in range(nb_frame):
proj_mat = projection_matrices[k]
cam_point = proj_mat @ point
cam_point /= cam_point[2]
frame = cv2.imread(f'data/torus/torus{k+1:04}.png') frame = cv2.imread(f'data/torus/torus{k+1:04}.png')
cv2.circle(frame, (int(cam_point[0]), int(cam_point[1])), 2, (0, 0, 255)) proj_mat = projection_matrices[k]
cam_points = proj_mat @ points.T
cam_points /= cam_points[2,:]
cam_points = np.round(cam_points, 0).astype(np.int32)
visible = np.logical_and.reduce((0 <= cam_points[0,:], cam_points[0,:] < frame.shape[1], 0 <= cam_points[1,:], cam_points[1,:] < frame.shape[0]))
cam_points = cam_points[:,visible]
points = points[visible,:]
solid = np.invert(((frame[cam_points[1,:],cam_points[0,:]] == background)).all(1))
cam_points = cam_points[:,solid]
points = points[solid,:]
for k in range(nb_frame):
frame = cv2.imread(f'data/torus/torus{k+1:04}.png')
# frame = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
# frame = 255 * (frame == 18).astype(np.uint8)
# frame = cv2.filter2D(frame, -1, np.ones((5, 5)) / 25)
# frame = 255 * (frame > 255/2).astype(np.uint8)
proj_mat = projection_matrices[k]
cam_points = proj_mat @ points.T
cam_points /= cam_points[2,:]
cam_points = np.round(cam_points, 0).astype(np.int32)
for cam_point in cam_points.T:
cv2.circle(frame, (cam_point[0], cam_point[1]), 2, (255, 0, 0))
cv2.imshow('Frame', frame) cv2.imshow('Frame', frame)
cv2.waitKey(0) cv2.waitKey(0)

Binary file not shown.