-
Notifications
You must be signed in to change notification settings - Fork 0
/
utilities.py
854 lines (684 loc) · 25.8 KB
/
utilities.py
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
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
#!/usr/bin/python3
import cv2
import numpy as np
import math
import json
import os
import time
import random
import sys
import datetime
from image import Image
def get_middle(img):
"""
Get the middle of the image
Res: P: (x,y)
"""
mid_x = int(round(img.shape[1] / 2, 0))
mid_y = int(round(img.shape[0] / 2, 0))
return (mid_x, mid_y)
def draw_label(to_show, text):
"""
Draw a label with defaults (position, etc.) that should work in most cases
"""
x_pos, y_pos = 20, 30
padding = 2
font_face = cv2.FONT_HERSHEY_SIMPLEX
font_scale = 1
thickness = 1
text_size, _ = cv2.getTextSize(text, font_face, font_scale, thickness)
cv2.rectangle(to_show, (x_pos-padding, y_pos-text_size[1]-padding), (x_pos+text_size[0]+padding, y_pos+padding), (0, 0, 0), cv2.FILLED)
cv2.putText(to_show, text, (x_pos, y_pos), font_face, font_scale, (255, 255, 255), thickness)
global_shown_history = []
global_show_i = 0
def show(img, win_name="test", fullscreen=False, time_ms=0, text=None, draw_histograms=False, keypoints=None, scale=True, keep=True):
"""
Show img in a window
"""
global global_show_i
if fullscreen:
cv2.namedWindow(win_name, cv2.WINDOW_NORMAL)
cv2.setWindowProperty(win_name, cv2.WND_PROP_FULLSCREEN, cv2.WINDOW_FULLSCREEN)
elif scale:
cv2.namedWindow(win_name, cv2.WINDOW_NORMAL|cv2.WINDOW_KEEPRATIO)
to_show = img.copy()
to_show = as_uint8(to_show)
if keypoints is not None:
to_show = cv2.drawKeypoints(to_show, keypoints, None,
flags=cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS)
if text is not None:
draw_label(to_show, text)
if draw_histograms:
n_channels = 0 if len(to_show.shape) < 3 else to_show.shape[2]
x_pos, y_pos = 10, 50
y_padding = 20
if n_channels == 0:
hist = draw_histogram(to_show)
to_show[y_pos:y_pos+hist.shape[0], x_pos:x_pos+hist.shape[1]] = hist
else:
for i in range(n_channels):
channel = to_show[:,:,i]
hist = draw_histogram(channel)
for j in range(n_channels):
to_show[y_pos:y_pos+hist.shape[0], x_pos:x_pos+hist.shape[1], j] = hist
y_pos += hist.shape[1] + y_padding
cv2.imshow(win_name, to_show)
if keep:
global_shown_history.append(to_show)
global_show_i = len(global_shown_history)-1
while True:
key = cv2.waitKey(time_ms)
char_key = key % 256
if char_key == ord('q'):
exit(0)
if char_key == ord('w'):
now = datetime.datetime.now().strftime("%Y-%m-%d_%H:%M:%S.png")
cv2.imwrite("imshow_"+now, to_show)
continue
elif char_key == ord('a'):
return show_all(Image(image_data=img))
elif char_key == ord('n'):
if global_show_i < len(global_shown_history)-1:
global_show_i += 1
return show(global_shown_history[global_show_i], keep=False)
else:
continue
elif char_key == ord('p'):
if global_show_i > 0:
global_show_i -= 1
return show(global_shown_history[global_show_i], keep=False)
else:
continue
if time_ms > 0:
break
if key not in (1114091, 1114089, 65515, 65513): # <win>, <alt> (used to move/resize windows)
break
#cv2.destroyWindow(win_name)
return chr(key%256)
def get_box(img, center=None, side=100):
"""
P: (x,y)
Get a square with side lengths side centered on center
"""
if center is None:
center = get_middle(img)
first_side = int(round(side / 2, 0))
second_side = side - first_side
return img[center[1]-first_side:center[1]+second_side, center[0]-first_side:center[0]+second_side]
def get_angle(p1, c, p2):
"""
P: (x,y)
Get the angle between the lines c -> p1 and c -> p2 in degrees
Treats the lines as being directionless, so the result will always be 0 <= angle <= 180
"""
line_1 = (p1[0]-c[0], p1[1]-c[1], 0)
line_2 = (p2[0]-c[0], p2[1]-c[1], 0)
v1_u = line_1 / np.linalg.norm(line_1)
v2_u = line_2 / np.linalg.norm(line_2)
angle_radians = np.arccos(np.clip(np.dot(v1_u, v2_u), -1.0, 1.0))
return angle_radians * 180 / math.pi
def flood_fill_until(img, limit, center=None, color=255, max_diff=100):
"""
Increase the loDiff and upDiff incrementally until limit of the pixels in the image has been filled
"""
seed = center
if seed is None:
seed = get_middle(img)
diff = 0
num_filled = 0
mask_shape = (img.shape[0]+2, img.shape[1]+2)
while num_filled < limit*img.size:# and diff < max_diff:
mask = np.zeros(mask_shape).astype(np.uint8)
num_filled, _, _, _ = cv2.floodFill(img, mask, seed, color, upDiff=diff, loDiff=diff, flags=cv2.FLOODFILL_FIXED_RANGE | cv2.FLOODFILL_MASK_ONLY | (color << 8))
diff += 1
# Remove the excess pixels around
mask = mask[1:-1,1:-1]
return mask
def draw_convex_hull(img, convex_hull):
copy = img.copy()
for idx, pt1 in enumerate(convex_hull):
if idx % 2 == 0:
color = (0, 0, 255)
else:
color = (255, 0, 0)
pt1 = pt1[0]
pt1 = (pt1[0], pt1[1])
pt2 = convex_hull[(idx + 1) % len(convex_hull)][0]
pt2 = (pt2[0], pt2[1])
cv2.line(copy, pt1, pt2, color, 3)
return copy
def poly2mask(poly, size_or_img):
size = size_or_img if type(size_or_img) != np.ndarray else size_or_img.shape[0:2]
mask = np.zeros(size, dtype=np.uint8)
if type(poly) == list:
poly = np.array(poly)
cv2.fillPoly(mask, [poly], 255)
return mask
def wait_for_key(char=None):
while(True):
key_code_raw = cv2.waitKey()
# http://stackoverflow.com/a/17284668/1517969
key_code = key_code_raw % 256
if char is None:
return chr(key_code)
elif key_code == ord(char):
return chr(key_code)
def select_polygon(orig_img):
"""
Interactively select a polygon. Add points with LB and finish with key "s"
"""
polygon = []
color = (0,0,255)
def mouse_callback(ev, x, y, flags, param):
img = orig_img.copy()
if ev == cv2.EVENT_LBUTTONDOWN:
polygon.append((x,y))
for p1,p2 in zip(polygon, polygon[1:]):
cv2.line(img, p1, p2, color, 2)
if len(polygon) > 0:
cv2.circle(img, polygon[0], 4, color)
cv2.line(img, polygon[-1], (x,y), color, 1)
if len(polygon) > 1:
cv2.line(img, (x,y), polygon[0], color, 1)
cv2.imshow("polygon-select", img)
cv2.namedWindow("polygon-select")
cv2.setMouseCallback("polygon-select", mouse_callback)
cv2.imshow("polygon-select", orig_img)
wait_for_key('s')
cv2.destroyWindow("polygon-select")
return polygon
def select_circles(img):
"""
Interactively select a number of circles. Add center with LB and approx. radius with
LB again. Then adjust position and radius with the keyboard.
Return a list of the circles. Each represented by a tuple: ((cx, cy), r)
"""
orig_img = img.copy()
canvas = orig_img.copy()
circles = []
center = None
color = (0,0,255)
# Make drag image (in qt viewer) work without triggering clicks:
mouse_moved_flag = False
message = ""
def draw(canvas, center, r):
cv2.circle(canvas, center, r, color, 1)
def redraw(canvas):
draw_label(canvas, message)
for center, r in circles:
draw(canvas, center, r)
return canvas
def mouse_callback(ev, x, y, flags, param):
nonlocal canvas
nonlocal center
nonlocal mouse_moved_flag
canvas = redraw(orig_img.copy())
r = round(math.sqrt((center[0]-x)**2 + (center[1]-y)**2) if center else 0)
if ev == cv2.EVENT_LBUTTONDOWN:
mouse_moved_flag = False
elif ev == cv2.EVENT_LBUTTONUP and not mouse_moved_flag:
if not center:
center = (x,y)
else:
circles.append((center, r))
draw(canvas, center, r)
center = None
elif ev == cv2.EVENT_MOUSEMOVE:
mouse_moved_flag = True
if not center:
pass
else:
draw(canvas, center, r)
cv2.imshow("circles-select", canvas)
cv2.namedWindow("circles-select")
cv2.setMouseCallback("circles-select", mouse_callback)
cv2.imshow("circles-select", orig_img)
# Over-engineering ftw :) (This might make sense for arbitrary rectange selections though)
state_labels = ['Move (jkli)', 'Grow/Shrink (jl)']
state_transforms = [
lambda x,y,r,dx,dy: ((x+dx, y+dy), r),
lambda x,y,r,dr,_: ((x, y), r+dr),
]
state = 0
active_transform = state_transforms[state]
# message = state_labels[state]
message = 'Move (jkli), Grow/Shrink (ed), Done (s), Quit (q)'
while True:
key = wait_for_key()
if key == 'u':
circles.pop()
elif key == 'n' and False:
state = (state + 1) % len(state_labels)
active_transform = state_transforms[state]
message = state_labels[state]
elif key == 's':
break
elif key == 'q':
exit(0)
if len(circles) > 0:
cur = circles[-1]
(x,y), r = cur
new_circle = None
if key == 'j':
new_circle = active_transform(x, y ,r , -1, 0)
elif key == 'l':
new_circle = active_transform(x, y ,r , 1, 0)
elif key == 'i':
new_circle = active_transform(x, y ,r , 0, -1)
elif key == 'k':
new_circle = active_transform(x, y ,r , 0, 1)
elif key == 'e':
new_circle = (cur[0],cur[1]+1) # radius
elif key == 'd':
new_circle = (cur[0],cur[1]-1) # radius
if new_circle:
circles[-1] = new_circle
cv2.imshow("circles-select", redraw(orig_img.copy()))
cv2.destroyWindow("circles-select")
return circles
def two_point_rect_to_bb(p1, p2):
"""
Converts a rectangle represented by two points to a bounding box: a (x,y,w,h) tuple
1----+ 2----o +----2 +----1 +-w--*
| | or | | or | | or | | -> h |
+----2 o----1 1----* 2----* (x,y)--*
"""
x = min(p1[0], p2[0])
y = min(p1[1], p2[1])
w = abs(p1[0] - p2[0])
h = abs(p1[1] - p2[1])
return (x, y, w, h)
from matplotlib import pyplot as plt
def plot_histogram(img, channels=[0], mask=None, colors=["b", "g", "r"], max=None):
"""
Adds the histogram to the active matplotlib plot. Use plt.show() after to show the plot.
"""
if(type(colors) == str):
colors=[colors]
max = np.max(img)+0.00001 if max is None else max
for idx, ch in enumerate(channels):
hist = cv2.calcHist([img], [ch], mask, [256], [0, max])
hist = hist/sum(hist) # normalize so each bucket represents percentage of total pixels
plt.plot(hist, colors[idx])
def get_histogram(single_channel_img):
single_channel_img = as_uint8(single_channel_img)
return cv2.calcHist([single_channel_img], [0], None, [256], [0, 256])
def draw_histogram(single_channel_img, max_height=256, padding=2,
ignored_values=[], fg=0, bg=255, mask=None):
"""
Get an image of the histogram (to be painted onto other images etc)
"""
single_channel_img = as_uint8(single_channel_img)
hist = cv2.calcHist([single_channel_img], [0], mask, [256], [0, 256])
value_type = np.uint8
if type(fg) != int or type(bg) != int:
value_type = np.float32
hist[ignored_values] = 0
hist_img = np.ones((max_height+2*padding, 256 + 2*padding), dtype=value_type)*bg
hist /= np.amax(hist)
hist *= max_height
for i in range(256):
x = i
y = int(hist[i][0])
pt1 = (padding+x, padding + max_height-y)
pt2 = (padding+x, padding + max_height-1)
cv2.line(hist_img, pt1, pt2, (fg))
return hist_img
def distance_point_to_line(line_pt1, line_pt2, pt):
"""
Get the distance from pt to the line passing through line_pt1 and line_pt2
"""
line_length = distance(line_pt1, line_pt2)
if line_length == 0:
return distance(line_pt1, pt)
return abs((line_pt2[1]-line_pt1[1])*pt[0] - (line_pt2[0] - line_pt1[0])*pt[1] + line_pt2[0]*line_pt1[1] - line_pt2[1]*line_pt1[0])\
/ line_length
def distance_point_to_bounded_line(line_pt1, line_pt2, pt):
"""
Get the distance from pt to the line passing through line_pt1 and line_pt2.
The line is bounded at line_pt1 and line_pt2, so if the closest point is "outside" the line, the distance to the
nearest point is returned.
Example:
line_pt2
o
/
/ pt
/ o
/
/
o
line_pt1
In this case the same number as returned from distance_point_to_line is returned
pt
o
line_pt2
o
/
/
/
/
/
o
line_pt1
In this case the euclidean distance from pt to line_pt2 is returned
"""
dist_infinite_line = distance_point_to_line(line_pt1, line_pt2, pt)
euclidean_dist_to_line_pt1 = distance(line_pt1, pt)
euclidean_dist_to_line_pt2 = distance(line_pt2, pt)
distance_along_line_to_line_pt1 = math.sqrt(euclidean_dist_to_line_pt1**2 - dist_infinite_line**2)
distance_along_line_to_line_pt2 = math.sqrt(euclidean_dist_to_line_pt2**2 - dist_infinite_line**2)
line_length = distance(line_pt1, line_pt2)
if distance_along_line_to_line_pt1 > line_length or distance_along_line_to_line_pt2 > line_length:
return min(distance_along_line_to_line_pt1, distance_along_line_to_line_pt2)
return dist_infinite_line
def get_metadata_path(img_path):
img_name = os.path.basename(img_path)
img_dir = img_path
is_dir_path = img_name.find(".") < 0
if not is_dir_path:
img_base = "".join(img_name.split(".")[:-1])
img_dir = os.path.dirname(img_path)
series_metadata_path = os.path.join(img_dir, "metadata.json")
if is_dir_path or os.path.exists(series_metadata_path):
return series_metadata_path
else:
return os.path.join(img_dir, img_base+".json")
def read_metadata(img_path):
metadata_path = get_metadata_path(img_path)
if os.path.exists(metadata_path):
with open(metadata_path) as fp:
meta_dict = json.load(fp)
return meta_dict
else:
return {}
def update_metadata(img_path, new_meta_data):
meta_dict = read_metadata(img_path)
meta_dict.update(new_meta_data)
metadata_path = get_metadata_path(img_path)
with open(metadata_path, "w") as fp:
json.dump(meta_dict, fp) # overwrites on error too...
return meta_dict
def transform_image(image, vec):
"""
:param image: Image object
:param vec: dict where keys = 1D color space (with preceding image space name: bgr_b, bgr_g etc.), and the value
is the coefficient to multiply that color space by
"""
transformed = None
for image_space_name, image_space_data in image.get_color_space_dict().items():
if transformed is None:
transformed = np.zeros(image_space_data.shape[:2])
if image_space_name in vec:
transformed += vec[image_space_name] * image_space_data
# Normalization
res = transformed - np.amin(transformed)
res /= np.amax(res)
return as_float32(res)
def as_uint8(img):
img = img.copy()
if img.dtype == np.uint8:
return img
if img.dtype not in (np.float32, np.float64):
raise RuntimeError("Unknown dtype: {}".format(img.dtype))
img -= np.amin(img)
img *= 1/np.amax(img)
img *= 255
img = np.around(img)
return img.astype(np.uint8)
def as_float32(img):
img = img.copy()
if img.dtype in (np.float32, np.float64):
return img.astype(np.float32) / np.amax(img)
if img.dtype != np.uint8:
raise RuntimeError("Unknown dtype: {}".format(img.dtype))
img = img.astype(np.float32)
return img / 255
def astype(img, dtype):
if dtype == np.float32:
return as_float32(img)
if dtype == np.uint8:
return as_uint8(img)
raise RuntimeError("Unknown dtype: {}".format(dtype))
class Timer:
def __init__(self):
self.start = time.time()
def reset(self):
self.start = time.time()
def __str__(self):
return str(round(time.time() - self.start, 3)) + "s"
def get_project_directory():
"""
Get an absolute path to the projects root directory
"""
cur_dir = os.path.dirname(__file__)
while cur_dir != "/" and not os.path.exists(os.path.join(cur_dir, ".git")):
cur_dir = os.path.dirname(cur_dir)
if not os.path.exists(os.path.join(cur_dir, ".git")):
raise FileNotFoundError("Unable to locate project directory!")
return cur_dir
def locate_file(path):
"""
Attempt to locate the file path in the project.
The path may be partial, as in:
path = microsoft_cam/24h/south/2016-04-12_19:21:04.png
will result in
/home/anders/UNIK4690/project/images/microsoft_cam/24h/south/2016-04-12_19:21:04.png
being returned (on my computer).
path = 2016-04-12_19:21:04.png would yield the same result (as no file called 2016-04-12_19:21:04.png exists
in any directory below /home/anders/UNIK4690/project/ except for in /home/anders/UNIK4690/project/images/microsoft_cam/24h/south/)
If there are duplicates, the first encountered by os.walk is returned.
"""
project_dir = get_project_directory()
for root, dirs, files in os.walk(project_dir):
# Remove directories we know can't contain the image
dir_copy = list(dirs)
for dir in dir_copy:
if dir[0] == "." or dir == "__pycache__":
dirs.remove(dir)
# print("Looking for {} in {}".format(path, root))
if os.path.exists(os.path.join(root, path)):
path = os.path.join(root, path)
break
return path
def show_all(image, time_ms=0):
bgr = image.get_bgr()
hsv = image.get_hsv()
lab = image.get_lab()
ycrcb = image.get_ycrcb()
spaces = []
for i in range(3):
spaces.append(bgr[:,:,i])
for i in range(3):
spaces.append(hsv[:,:,i])
for i in range(3):
spaces.append(lab[:,:,i])
for i in range(3):
spaces.append(ycrcb[:,:,i])
to_show = np.zeros(bgr.shape[:2])
width_per = bgr.shape[1] // 4
height_per = bgr.shape[0] // 3
idx = 0
y = 0
for row in (1,2,3):
x = 0
for col in (1,2,3,4):
to_show[y:y+height_per, x:x+width_per] = cv2.resize(spaces[idx], (width_per, height_per), cv2.INTER_CUBIC)
idx += 1
x += width_per
y += height_per
return show(to_show, time_ms=time_ms, fullscreen=True)
def distance(pt1, pt2):
"""
Euclidean distance from pt1 to pt2
"""
import math
return math.sqrt((pt1[0]-pt2[0])**2 + (pt1[1]-pt2[1])**2)
def Timed(fn):
"""
Decorator to print duration of function calls
"""
def wrapped(*args, **kwargs):
timer = Timer()
ret = fn(*args, **kwargs)
print("%s took: %s" % (fn, timer))
return ret
return wrapped
# @Timed
def polygon_symmetric_diff(a, b):
"""
Returns the symmetric difference between the two polygons. Ie. pixel count not shared by both polygons.
NB: Probably assumes convex polygons
"""
# Brute force approach. Make masks for each polygon and count the pixels in the symmetric difference
x,y,w,h = cv2.boundingRect(np.concatenate((a,b)))
a = a - (x,y)
b = b - (x,y)
a_mask = poly2mask(a, (h,w))
b_mask = poly2mask(b, (h,w))
sym_diff = cv2.bitwise_xor(a_mask, b_mask)
return np.count_nonzero(sym_diff), (a_mask, b_mask)
def playground_score(known, detected):
diff, masks = polygon_symmetric_diff(known, detected)
known_area = masks[0]
return 1.0 - diff/np.count_nonzero(known_area)
def matching_balls(known, detected, match_threshold_factor=1.0):
"""
known: [(center, r), ...]
detected: [center, ...]
match_threshold_factor: a ball x is considered a match of y if |x-y| < x_r*thresh
We don't distinguish between ball types here.
In this version we simply go through the known ball, matching them with the
best match. No ball will be matched more than once.
This doesn't necessarily maximize the score though:
Should we attempt to find the optimal pairing?
Illustrative example:
*..O.*..O...*
a A b B c
Assuming the radius of both A and B is 3 (...) should we attempt to pair B with b
and A with a? Even though b match better with A.
Or should b match with both A and B?
Similarily: should A match with both a and b?
"""
# There's only a few balls so we don't do anything fancy here yet
matches = []
unmatched = list(detected)
for known_ball in known:
if len(unmatched) == 0:
break
center, r = known_ball
match_idx, best_match = min(enumerate(unmatched), key=lambda x: distance(center, x[1]))
d = distance(center, best_match)
if d < r*match_threshold_factor:
matches.append((known_ball, best_match))
del unmatched[match_idx]
return matches
def ball_detection_score(known, detected):
# Three kinda of errors: (1) ball not detected at all, (2) offset from hand-detected center, (3) non-ball detected
#
# Only consider error 1 and 3 here
#
# Muliple possible scoring criteria:
# - precision (same as accuracy in this case)
# - recall
# - F-score
matches = matching_balls(known, detected)
match_count = len(matches)
recall = match_count / len(known)
precision = match_count / len(detected)
if recall + precision < 0.000000001:
return 0, matches
# F-score, beta = 1:
return 2*recall*precision / (recall + precision), matches
def circle_bb(circle, margin=0):
"""
Bounding box of a circle represeted by ((cx,cy), r)
"""
(cx,cy), r = circle
r = r+margin
x, y = cx-r, cy-r
# How to interpret radius and center in a pixel world?
# Kinda makes sense to to force odd diameter, but seems like opencv don't do that.
# https://github.com/Itseez/opencv/blob/2f4e38c8313ff313de7c41141d56d945d91f47cf/modules/imgproc/src/drawing.cpp#L1411
# See test_circle_drawing
# (---+---) or (---+--)
w = r*2
h = w
return (x,y,w,h)
def extract_circle(img, circle, margin=0, mask_color=None):
(cx, cy), r = circle
x, y, h, w = circle_bb(((cx, cy), r+margin))
roi = img[y:y+h, x:x+w]
if mask_color:
mask = np.zeros((h, w), dtype=np.uint8)
cv2.circle(mask, (cx-x, cy-y), r, 255, -1)
roi[np.where(mask == 0)] = mask_color
return roi
def extract_bb(img, r):
x,y,w,h = r
return img[y:y+h, x:x+w]
def pretty_print_keypoint(kp):
return "<(%.0f, %.0f), %d, %.1f, %d>" % (kp.pt[0], kp.pt[1], kp.size, kp.response, kp.octave)
def power_threshold(float_img, exponent):
# Exponentiating the float_img channel (as a float image [0,1]) creates sort of a soft threshold.
# (Suppressing darker values)
float_img = np.power(float_img, exponent)
amax = np.amax(float_img)
light = float_img / amax
light *= 255
light = np.clip(light, 0, 255)
light = light.astype(np.uint8)
return light
def make_debug_toggable(fn, key=""):
import os
def wrapped(*args, **kwargs):
value = os.environ.get("DEBUG")
if value is None:
return
if key in value:
return fn(*args, **kwargs)
return wrapped
def keypoint_filter_overlapping(kps):
"""
The surf detector often finds multiple key points close to each other.
Here we simply looks naively for overlapping points reducing each "cluster"
to just one point. (Selecting the largest one)
This seems to work for some common cases at least.
"""
if len(kps) > 150:
print("Not dimensioned for huge number of key points O(N^2)", file=sys.stderr)
return kps
def overlapping(it, kps):
# NB: includes itself in result
overlaps = []
complement = []
for kp in kps:
dist = distance(it.pt, kp.pt)
# Seems kp.size is the diameter (ish at least.. it's a bit unclear..)
if dist < kp.size/2+it.size:
overlaps.append(kp)
else:
complement.append(kp)
return overlaps, complement
filtered = []
kps = sorted(kps, key=lambda kp: kp.size)
# We don't find transitive overlaps, but we do check the largest key points first
while len(kps) > 0:
kp = kps[-1]
cluster, kps = overlapping(kp, kps)
# Note the key points also have a 'response' field.
filtered.append(max(cluster, key=lambda kp: kp.size))
return filtered
if __name__ == "__main__":
img_paths = ["raw/1.jpg", "raw/2.jpg", "raw/3.jpg", "24h/south/latest.png", "24h/south/2016-04-12_18:59:03.png", "24h/south/2016-04-12_19:21:04.png", "24h/south/2016-04-13_09:03:03.png", "24h/south/2016-04-13_12:45:04.png"]
images = list(map(cv2.imread, ["images/microsoft_cam/"+img_path for img_path in img_paths]))
images = [i for i in images if i is not None]
# 90
print(get_angle((0,100), (0,0), (100,0)))
# 90
print(get_angle((0,-100), (0,0), (-100,0)))
# 90
print(get_angle((0,100), (100,100), (100,0)))
# ~0
print(get_angle((0,100), (-100,-100), (0,100)))
# ~180
print(get_angle((-100,100), (0,0), (100,-100)))