1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
|
import argparse
import collections
import sys
import cv2
import matplotlib.pyplot as plt
import numpy as np
from creedsolo import RSCodec
from raptorq import Decoder
parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument("-i", "--input", help="camera device index or input video file", default=0)
parser.add_argument("-o", "--output", help="output file for decoded data")
parser.add_argument("-x", "--height", help="grid height", default=100, type=int)
parser.add_argument("-y", "--width", help="grid width", default=100, type=int)
parser.add_argument("-f", "--fps", help="frame rate", default=30, type=int)
parser.add_argument("-l", "--level", help="error correction level", default=0.1, type=float)
parser.add_argument("-s", "--size", help="number of bytes to decode", default=2**16, type=int)
args = parser.parse_args()
cheight = cwidth = max(args.height // 10, args.width // 10)
frame_size = args.height * args.width - 4 * cheight * cwidth
frame_xor = np.arange(frame_size, dtype=np.uint8)
rs_size = frame_size - int((frame_size + 254) / 255) * int(args.level * 255) - 4
rsc = RSCodec(int(args.level * 255))
decoder = Decoder.with_defaults(args.size, rs_size)
if args.input.isdecimal():
args.input = int(args.input)
cap = cv2.VideoCapture(args.input)
data = None
while data is None:
try:
ret, raw_frame = cap.read()
if not ret:
print("End of stream")
sys.exit(1)
raw_frame = cv2.cvtColor(raw_frame, cv2.COLOR_BGR2RGB).astype(np.float64)
X, Y = raw_frame.shape[:2]
scale = min(X // 20, Y // 20)
# Resize so smaller dim is 20
# Use fast default interpolation for factor of 4
# Then switch to good slow interpolation
dframe = cv2.resize(
cv2.resize(raw_frame, (Y // 4, X // 4)),
(Y // scale, X // scale), # OpenCV swaps them
interpolation=cv2.INTER_AREA,
)
# plt.imshow(dframe.astype(np.uint8))
# plt.show()
def max_in_orig(x):
return tuple(np.array(np.unravel_index(np.argmax(x), x.shape)) * scale + scale // 2)
sumframe = np.sum(dframe, axis=2)
# TODO: Only search in corner area
widx = max_in_orig((np.std(dframe, axis=2) < 35) * sumframe)
ridx = max_in_orig(2 * dframe[:, :, 0] - sumframe)
gidx = max_in_orig(2 * dframe[:, :, 1] - sumframe)
bidx = max_in_orig(2 * dframe[:, :, 2] - sumframe)
# Flood fill corners
def flood_fill(s):
# TODO: make this faster
vis = np.full((X, Y), False)
vis[s] = True
queue = collections.deque([s])
pos = np.array(s)
col = np.copy(raw_frame[s])
n = 1
while len(queue) > 0:
u = queue.popleft()
for d in [(1, 0), (0, 1), (-1, 0), (0, -1)]:
v = (u[0] + d[0], u[1] + d[1])
if (
0 <= v[0] < X
and 0 <= v[1] < Y
and not vis[v]
and np.linalg.norm(raw_frame[v] - raw_frame[s]) < 125
):
vis[v] = True
pos += np.array(v)
col += raw_frame[v]
n += 1
queue.append(v)
# plt.imshow(raw_frame.astype(np.uint8))
# plt.scatter(*reversed(np.where(vis)))
# plt.scatter(pos[1] / n, pos[0] / n)
# plt.show()
return pos / n, col / n
widx, wcol = flood_fill(widx)
ridx, rcol = flood_fill(ridx)
gidx, gcol = flood_fill(gidx)
bidx, bcol = flood_fill(bidx)
# Find basis of color space
origin = (rcol + gcol + bcol - wcol) / 2
rcol -= origin
gcol -= origin
bcol -= origin
F = 255 * np.linalg.inv(np.stack((rcol, gcol, bcol)).T)
# Dumb perspective transform
xv = np.linspace(
-(cheight / 2 - 1) / (args.height - cheight + 1),
1 + (cheight / 2 - 1) / (args.height - cheight + 1),
args.height,
)
yv = np.linspace(
-(cwidth / 2 - 1) / (args.width - cwidth + 1),
1 + (cwidth / 2 - 1) / (args.width - cwidth + 1),
args.width,
)
xp = (
np.outer(1 - xv, 1 - yv) * widx[0]
+ np.outer(1 - xv, yv) * ridx[0]
+ np.outer(xv, 1 - yv) * gidx[0]
+ np.outer(xv, yv) * bidx[0]
)
yp = (
np.outer(1 - xv, 1 - yv) * widx[1]
+ np.outer(1 - xv, yv) * ridx[1]
+ np.outer(xv, 1 - yv) * gidx[1]
+ np.outer(xv, yv) * bidx[1]
)
# plt.scatter(widx[1], widx[0])
# plt.scatter(ridx[1], ridx[0])
# plt.scatter(gidx[1], gidx[0])
# plt.scatter(bidx[1], bidx[0])
# plt.scatter(yp, xp)
# plt.imshow(raw_frame.astype(np.uint8))
# plt.show()
raw_color_frame = raw_frame[np.round(xp).astype(np.int64), np.round(yp).astype(np.int64), :]
# color_frame = raw_color_frame.astype(np.uint8)
color_frame = np.clip(np.squeeze(F @ (raw_color_frame - origin)[..., np.newaxis]), 0, 255).astype(np.uint8)
frame = (
(color_frame[:, :, 0] >> 5) + (color_frame[:, :, 1] >> 2 & 0b00111000) + (color_frame[:, :, 2] & 0b11000000)
)
frame_data = np.concatenate(
(
frame[:cheight, cwidth : args.width - cwidth].flatten(),
frame[cheight : args.height - cheight].flatten(),
frame[args.height - cheight :, cwidth : args.width - cwidth].flatten(),
)
)
data = decoder.decode(bytes(rsc.decode(frame_data ^ frame_xor)[0]))
print("Decoded frame")
except Exception as e:
print(e)
with open(args.output, "wb") as f:
f.write(data)
cap.release()
|
