-
Notifications
You must be signed in to change notification settings - Fork 29
/
eventvision.py
691 lines (571 loc) · 28 KB
/
eventvision.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
# -*- coding: utf-8 -*-
"""
This module contains classes, functions and an example (main) for handling AER vision data.
"""
import glob
import cv2
import numpy as np
from win32api import GetSystemMetrics
import timer
class Events(object):
"""
Temporal Difference events.
data: a NumPy Record Array with the following named fields
x: pixel x coordinate, unsigned 16bit int
y: pixel y coordinate, unsigned 16bit int
p: polarity value, boolean. False=off, True=on
ts: timestamp in microseconds, unsigned 64bit int
width: The width of the frame. Default = 304.
height: The height of the frame. Default = 240.
"""
def __init__(self, num_events, width=304, height=240):
"""num_spikes: number of events this instance will initially contain"""
self.data = np.rec.array(None, dtype=[('x', np.uint16), ('y', np.uint16), ('p', np.bool_), ('ts', np.uint64)], shape=(num_events))
self.width = width
self.height = height
def show_em(self):
"""Displays the EM events (grayscale ATIS events)"""
frame_length = 24e3
t_max = self.data.ts[-1]
frame_start = self.data[0].ts
frame_end = self.data[0].ts + frame_length
max_val = 1.16e5
min_val = 1.74e3
val_range = max_val - min_val
thr = np.rec.array(None, dtype=[('valid', np.bool_), ('low', np.uint64), ('high', np.uint64)], shape=(self.height, self.width))
thr.valid.fill(False)
thr.low.fill(frame_start)
thr.high.fill(0)
def show_em_frame(frame_data):
"""Prepare and show a single frame of em data to be shown"""
for datum in np.nditer(frame_data):
ts_val = datum['ts'].item(0)
thr_data = thr[datum['y'].item(0), datum['x'].item(0)]
if datum['p'].item(0) == 0:
thr_data.valid = 1
thr_data.low = ts_val
elif thr_data.valid == 1:
thr_data.valid = 0
thr_data.high = ts_val - thr_data.low
img = 255 * (1 - (thr.high - min_val) / (val_range))
#thr_h = cv2.adaptiveThreshold(thr_h, 255,
#cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 3, 0)
img = np.piecewise(img, [img <= 0, (img > 0) & (img < 255), img >= 255], [0, lambda x: x, 255])
img = img.astype('uint8')
cv2.imshow('img', img)
cv2.waitKey(1)
while frame_start < t_max:
#with timer.Timer() as em_playback_timer:
frame_data = self.data[(self.data.ts >= frame_start) & (self.data.ts < frame_end)]
show_em_frame(frame_data)
frame_start = frame_end + 1
frame_end += frame_length + 1
#print 'showing em frame took %s seconds' %em_playback_timer.secs
cv2.destroyAllWindows()
return
def show_td(self, wait_delay=1):
"""Displays the TD events (change detection ATIS or DVS events)
waitDelay: milliseconds
"""
frame_length = 24e3
t_max = self.data.ts[-1]
frame_start = self.data[0].ts
frame_end = self.data[0].ts + frame_length
td_img = np.ones((self.height, self.width), dtype=np.uint8)
while frame_start < t_max:
frame_data = self.data[(self.data.ts >= frame_start) & (self.data.ts < frame_end)]
if frame_data.size > 0:
td_img.fill(128)
#with timer.Timer() as em_playback_timer:
for datum in np.nditer(frame_data):
td_img[datum['y'].item(0), datum['x'].item(0)] = datum['p'].item(0)
#print 'prepare td frame by iterating events took %s seconds'
#%em_playback_timer.secs
td_img = np.piecewise(td_img, [td_img == 0, td_img == 1, td_img == 128], [0, 255, 128])
cv2.imshow('img', td_img)
cv2.waitKey(wait_delay)
frame_start = frame_end + 1
frame_end = frame_end + frame_length + 1
cv2.destroyAllWindows()
return
def filter_td(self, us_time):
"""Generate a filtered set of event data.
Does not modify instance data
Uses a background activity filter on the events, such that only events which are
correlated with a neighbouring event within 'us_time' microseconds will be allowed
through the filter.
us_time: microseconds
"""
max_x = self.width - 1
max_y = self.height - 1
t0 = np.ones((self.width, self.height)) - us_time - 1
x_prev = 0
y_prev = 0
p_prev = 0
valid_indices = np.ones(len(self.data), np.bool_)
i = 0
with timer.Timer() as ref_timer:
for datum in np.nditer(self.data):
datum_ts = datum['ts'].item(0)
datum_x = datum['x'].item(0)
datum_y = datum['y'].item(0)
datum_p = datum['p'].item(0)
if x_prev != datum_x | y_prev != datum_y | p_prev != datum_p:
t0[datum_x, datum_y] = -us_time
min_x_sub = max(0, datum_x - 1)
max_x_sub = min(max_x, datum_x + 1)
min_y_sub = max(0, datum_y - 1)
max_y_sub = min(max_y, datum_y + 1)
t0_temp = t0[min_x_sub:(max_x_sub + 1), min_y_sub:(max_y_sub + 1)]
if min(datum_ts - t0_temp.reshape(-1, 1)) > us_time:
valid_indices[i] = 0
t0[datum_x, datum_y] = datum_ts
x_prev = datum_x
y_prev = datum_y
p_prev = datum_p
i = i + 1
print 'filtering took %s seconds' % ref_timer.secs
return self.data[valid_indices.astype('bool')]
def sort_order(self):
"""Generate data sorted by ascending ts
Does not modify instance data
Will look through the struct events, and sort all events by the field 'ts'.
In other words, it will ensure events_out.ts is monotonically increasing,
which is useful when combining events from multiple recordings.
"""
#chose mergesort because it is a stable sort, at the expense of more
#memory usage
events_out = np.sort(self.data, order='ts', kind='mergesort')
return events_out
def extract_roi(self, top_left, size, is_normalize=False):
"""Extract Region of Interest
Does not modify instance data
Generates a set of td_events which fall into a rectangular region of interest with
top left corner at 'top_left' and size 'size'
top_left: [x: int, y: int]
size: [width, height]
is_normalize: bool. If True, x and y values will be normalized to the cropped region
"""
min_x = top_left[0]
min_y = top_left[1]
max_x = size[0] + min_x
max_y = size[1] + min_y
extracted_data = self.data[(self.data.x >= min_x) & (self.data.x < max_x) & (self.data.y >= min_y) & (self.data.y < max_y)]
if is_normalize:
self.width = size[0]
self.height = size[1]
extracted_data = np.copy(extracted_data)
extracted_data = extracted_data.view(np.recarray)
extracted_data.x -= min_x
extracted_data.y -= min_y
return extracted_data
def apply_refraction(self, us_time):
"""Implements a refractory period for each pixel.
Does not modify instance data
In other words, if an event occurs within 'us_time' microseconds of
a previous event at the same pixel, then the second event is removed
us_time: time in microseconds
"""
t0 = np.ones((self.width, self.height)) - us_time - 1
valid_indices = np.ones(len(self.data), np.bool_)
#with timer.Timer() as ref_timer:
i = 0
for datum in np.nditer(self.data):
datum_ts = datum['ts'].item(0)
datum_x = datum['x'].item(0)
datum_y = datum['y'].item(0)
if datum_ts - t0[datum_x, datum_y] < us_time:
valid_indices[i] = 0
else:
t0[datum_x, datum_y] = datum_ts
i += 1
#print 'Refraction took %s seconds' % ref_timer.secs
return self.data[valid_indices.astype('bool')]
def write_j_aer(self, filename):
"""
writes the td events in 'td_events' to a file specified by 'filename'
which is compatible with the jAER framework.
To view these events in jAER, make sure to select the DAVIS640 sensor.
"""
import time
y = 479 - self.data.y
#y = td_events.y
y_shift = 22 + 32
x = 639 - self.data.x
#x = td_events.x
x_shift = 12 + 32
p = self.data.p + 1
p_shift = 11 + 32
ts_shift = 0
y_final = y.astype(dtype=np.uint64) << y_shift
x_final = x.astype(dtype=np.uint64) << x_shift
p_final = p.astype(dtype=np.uint64) << p_shift
ts_final = self.data.ts.astype(dtype=np.uint64) << ts_shift
vector_all = np.array(y_final + x_final + p_final + ts_final, dtype=np.uint64)
aedat_file = open(filename, 'wb')
version = '2.0'
aedat_file.write('#!AER-DAT' + version + '\r\n')
aedat_file.write('# This is a raw AE data file - do not edit\r\n')
aedat_file.write \
('# Data format is int32 address, int32 timestamp (8 bytes total), repeated for each event\r\n')
aedat_file.write('# Timestamps tick is 1 us\r\n')
aedat_file.write('# created ' + time.strftime("%d/%m/%Y") \
+ ' ' + time.strftime("%H:%M:%S") \
+ ' by the Python function "write2jAER"\r\n')
aedat_file.write \
('# This function fakes the format of DAVIS640 to allow for the full ATIS address space to be used (304x240)\r\n')
##aedat_file.write(vector_all.astype(dtype='>u8').tostring())
to_write = bytearray(vector_all[::-1])
to_write.reverse()
aedat_file.write(to_write)
#aedat_file.write(vector_all)
#vector_all.tofile(aedat_file)
aedat_file.close()
def present_checkerboard(num_squares):
"""
Presents a checkerboard pattern of size num_squares*num_squares on the screen.
The function will automatically detect the screen size in pixels and assume a
resolution of 96 dpi to provide the square size in mm.
"""
screen_width_pixels = GetSystemMetrics(0)
screen_height_pixels = GetSystemMetrics(1)
#fixed parameters of the setup
figure_border_size = 30 #leave space of 100 pixels on each side of the axes for the figure
#controls etc
#image_border_size = 10 #within the image, create a border of size 10
#pixels to ensure contrast with the outside
#rectangles
#How big is each rectangle in units of pixels?
screen_size_pixels = np.array([screen_width_pixels, screen_height_pixels])
screen_size_mm = 0.00254 * screen_size_pixels / 96
square_size_pixels = int(min(screen_size_pixels - 2 * figure_border_size) / (num_squares + 2))
image_border_size = np.array([1, 2])
image_border_size[0] = (screen_size_pixels[0] - figure_border_size * 2 - square_size_pixels * (num_squares)) / 2
image_border_size[1] = (screen_size_pixels[1] - figure_border_size * 2 - square_size_pixels * (num_squares)) / 2
#How big is each rectangle in units of millimeters?
square_size_mm = screen_size_mm * square_size_pixels / screen_size_pixels
#How big is the checkered part of the image
image_inner_dim = num_squares * square_size_pixels # the dimenstion of the inside of the image (not including the border)
#Create a black image to fit both the checkerboard and the image border
img_template = np.ones((image_inner_dim + 2 * image_border_size[1], image_inner_dim + 2 * image_border_size[0]))
## create the checkerboard image
img = img_template
for x in range(0, num_squares):
for y in range((x) % 2, num_squares, 2):
minx = image_border_size[1] + (x) * square_size_pixels
maxx = image_border_size[1] + (x + 1) * square_size_pixels
miny = image_border_size[0] + (y) * square_size_pixels
maxy = image_border_size[0] + (y + 1) * square_size_pixels
img[minx:maxx, miny:maxy] = 1
for y in range((x + 1) % 2, num_squares, 2):
minx = image_border_size[1] + (x) * square_size_pixels
maxx = image_border_size[1] + (x + 1) * square_size_pixels
miny = image_border_size[0] + (y) * square_size_pixels
maxy = image_border_size[0] + (y + 1) * square_size_pixels
img[minx:maxx, miny:maxy] = 0
#xloc =
#range(image_borderSize+((x-1)*squareSize_pixels),(x*squareSize_pixels+image_borderSize))
#yloc =
#range(image_borderSize+((y-1)*squareSize_pixels),(y*squareSize_pixels+image_borderSize))
#img[[xloc],[yloc]] = 0
# display
cv2.imshow('image', img)
print 'Warning: Do not resize the checkerboard image window! It has been shown on the screen at a specific size which must be known for calibration'
print 'press any key when done recording images'
cv2.waitKey(0)
cv2.destroyAllWindows()
#print('Checkerboard rectangle size is:')
#print(['Vertical: ', num2str(squareSize_mm(2)), 'mm'])
#print(['Horizontal: ', num2str(squareSize_mm(1)), 'mm'])
#if num_flashes>1
# print('Press any button to begin flashing...\n')
# cv2.Waitkey(0)
# cv2.imshow('image', img)
# pause(1) %small pause
#
# % flash 'num_flashes' times
# for i = 1:num_flashes
# imshow(imgTemplate')
# drawnow
# imshow(img')
# drawnow
# end
#end
#
#dX = squareSize_mm(1)
#dY = squareSize_mm(2)
return square_size_mm
def auto_calibrate(num_squares, square_size_mm, scale, image_directory, image_format):
"""
Will read in images of extension 'image_format' from the directory 'image_directory' for calibration.
Each image should contain a checkerboard with 'num_squares'*'num_squares' squares,
each of size 'squareSize_mm'.
'scale' is an optional argument to rescale images before calibration
because ATIS/DVS have very low resolution and calibration algorithms are used to handling larger images (use 4)
"""
# termination criteria
criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 30, 0.001)
# how large are the squares?
square_sidelength = square_size_mm[1]
# prepare object points, like (0,0,0), (1,0,0), (2,0,0) ....,(6,5,0)
objp = np.zeros(((num_squares - 1) * (num_squares - 1), 3), np.float32)
objp[:, :2] = np.mgrid[0:(num_squares - 1), 0:(num_squares - 1)].T.reshape(-1, 2) * square_sidelength
# Arrays to store object points and image points from all the images.
objpoints = [] # 3d point in real world space
imgpoints = [] # 2d points in image plane.
images = glob.glob(image_directory + '\\*' + image_format)
for fname in images:
img_original = cv2.imread(fname)
gray_original = cv2.cvtColor(img_original, cv2.COLOR_BGR2GRAY)
#img = cv2.resize(img_small, None, fx=1, fy=1, interpolation =
#cv2.INTER_CUBIC)
img = cv2.resize(img_original, None, fx=scale, fy=scale, interpolation=cv2.INTER_CUBIC)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
#cv2.imshow('img',gray)
#cv2.waitKey(0)
#gray = cv2.equalizeHist(gray)
threshold = 128
keypressed = 0
while keypressed != 13:
ret, gray_threshold = cv2.threshold(gray, threshold, 255, cv2.THRESH_BINARY_INV)
cv2.imshow('img', gray_threshold)
keypressed = cv2.waitKey(0)
if keypressed == 2490368:
threshold = threshold + 1
if keypressed == 2621440:
threshold = threshold - 1
gray = gray_threshold
# Find the chess board corners
#ret, corners = cv2.findChessboardCorners(gray, (9,9),
#flags=cv2.CALIB_CB_ADAPTIVE_THRESH+cv2.CALIB_CB_NORMALIZE_IMAGE)
ret, corners = cv2.findChessboardCorners(gray, ((num_squares - 1), (num_squares - 1)), flags=cv2.CALIB_CB_NORMALIZE_IMAGE)
# If found, add object points, image points (after refining them)
if ret:
#if not (corners is None):
objpoints.append(objp)
corners2 = cv2.cornerSubPix(gray, corners, (4, 4), (-1, -1), criteria)
imgpoints.append(corners2 / scale)
# Draw and display the corners
img = cv2.drawChessboardCorners(img, ((num_squares - 1), (num_squares - 1)), corners2, ret)
cv2.imshow('img', img)
cv2.waitKey(0)
cv2.destroyAllWindows()
# perform the calibration
ret, mtx, dist, rvecs, tvecs = cv2.calibrateCamera(objpoints, imgpoints, gray_original.shape[::-1], None, None)
# calculate the error
tot_error = 0
for i in xrange(len(objpoints)):
imgpoints2, _ = cv2.projectPoints(objpoints[i], rvecs[i], tvecs[i], mtx, dist)
error = cv2.norm(imgpoints[i], imgpoints2, cv2.NORM_L2) / len(imgpoints2)
tot_error += error
print "mean error: ", tot_error / len(objpoints)
return ret, mtx, dist, rvecs, tvecs
def read_aer(filename):
"""Reads in the ATIS file specified by 'filename' and returns the TD and EM events.
This only works for ATIS recordings directly from the GUI.
If you are working with the N-MNIST or N-CALTECH101 datasets, use read_dataset(filename) instead
"""
with timer.Timer() as read_aer_timer:
file_handle = open(filename, 'rb')
#raw_data = np.fromfile(f, dtype=np.uint8, count=-1)
raw_data = np.fromfile(file_handle, dtype=np.uint8)
file_handle.close()
print '=> Reading .val file took %s s' % read_aer_timer.secs
with timer.Timer() as read_aer_timer:
raw_data = np.uint16(raw_data)
all_y = raw_data[3::4]
all_x = ((raw_data[1::4] & 32) << 3) | raw_data[2::4] #bit 5
all_p = (raw_data[1::4] & 128) >> 7 #bit 7
#all_ts = raw_data[0::4] | ((raw_data[1::4] & 31) << 8) # bit 4 downto
#0
all_ts2 = raw_data[0::4] | ((raw_data[1::4] & 31) << 8) # bit 4 downto 0
#all_event_type = (raw_data[1::4] & 64) >> 6 #bit 6
all_event_type2 = (raw_data[1::4] & 64) >> 6 #bit 6
#all_ts = all_ts.astype('uint')
all_ts2 = all_ts2.astype('uint')
print '=> Parsing .val data took %s s' % read_aer_timer.secs
time_increment = 2 ** 13
##old way, much slower
#with timer.Timer() as read_aer_timer:
# td_event_indices = np.zeros(len(all_y), dtype=np.bool_)
# em_event_indices = np.copy(td_event_indices)
# time_offset = 0
# for i, y_val in enumerate(all_y):
# if (y_val == 240) and (all_x[i] == 305):
# #timestamp overflow, increment the time offset
# time_offset += time_increment
# else:
# #apply time offset
# all_ts[i] += time_offset
# #update the td and em event indices
# em_event_indices[i] = all_event_type[i]
# td_event_indices[i] = not em_event_indices[i]
#print '=> Processing .val data old way took %s s' % read_aer_timer.secs
# Process time stamp overflow events,
# then generate the td and em event indices
with timer.Timer() as read_aer_timer:
overflow_indices = np.where(all_y == 240)[0]
for overflow_index in overflow_indices:
all_ts2[overflow_index:] += time_increment
all_event_type2[overflow_indices] = 2
em_event_indices2 = np.where(all_event_type2 == 1)[0]
td_event_indices2 = np.where(all_event_type2 == 0)[0]
print '=> Processing .val data new way took %s s' % read_aer_timer.secs
#em = Events(em_event_indices.sum())
#em.data.x = all_x[em_event_indices]
#em.data.y = all_y[em_event_indices]
#em.data.ts = all_ts[em_event_indices]
#em.data.p = all_p[em_event_indices]
#td = Events(td_event_indices.sum())
#td.data.x = all_x[td_event_indices]
#td.data.y = all_y[td_event_indices]
#td.data.ts = all_ts[td_event_indices]
#td.data.p = all_p[td_event_indices]
em2 = Events(em_event_indices2.size)
em2.data.x = all_x[em_event_indices2]
em2.data.y = all_y[em_event_indices2]
em2.data.ts = all_ts2[em_event_indices2]
em2.data.p = all_p[em_event_indices2]
#if datum.p == 0:
# thr_valid[datum.y, datum.x] = 1
# thr_l[datum.y, datum.x] = datum.ts
#elif thr_valid[datum.y, datum.x] == 1:
# thr_valid[datum.y, datum.x] = 0
# thr_h[datum.y, datum.x] = datum.ts - thr_l[datum.y, datum.x]
td2 = Events(td_event_indices2.size)
td2.data.x = all_x[td_event_indices2]
td2.data.y = all_y[td_event_indices2]
td2.data.ts = all_ts2[td_event_indices2]
td2.data.p = all_p[td_event_indices2]
##test correctness of new way
#print np.array_equal(em, em2)
#print np.array_equal(td, td2)
# It appears that the polarity needs to be flipped (when results are compared with Matlab output).
# Change the polarity: 0 events become 1 events and vice versa.
td2.data.p = np.abs(td2.data.p - 1)
return td2, em2
def read_dataset(filename):
"""Reads in the TD events contained in the N-MNIST/N-CALTECH101 dataset file specified by 'filename'"""
f = open(filename, 'rb')
raw_data = np.fromfile(f, dtype=np.uint8)
f.close()
raw_data = np.uint32(raw_data)
all_y = raw_data[1::5]
all_x = raw_data[0::5]
all_p = (raw_data[2::5] & 128) >> 7 #bit 7
all_ts = ((raw_data[2::5] & 127) << 16) | (raw_data[3::5] << 8) | (raw_data[4::5])
#Process time stamp overflow events
time_increment = 2 ** 13
overflow_indices = np.where(all_y == 240)[0]
for overflow_index in overflow_indices:
all_ts[overflow_index:] += time_increment
#Everything else is a proper td spike
td_indices = np.where(all_y != 240)[0]
td = Events(td_indices.size, 34, 34)
td.data.x = all_x[td_indices]
td.width = td.data.x.max() + 1
td.data.y = all_y[td_indices]
td.height = td.data.y.max() + 1
td.data.ts = all_ts[td_indices]
td.data.p = all_p[td_indices]
return td
def read_bin_linux(filename):
"""Reads in ATIS .bin files generated by the linux interface.
If working with N-MNIST or N-CALTECH101 datasets, use read_dataset(filename).
If working with recordings from the GUI, use read_aer(filename).
Returns TD, containing:
data: a NumPy Record Array with the following named fields
x: pixel x coordinate, unsigned 16bit int
y: pixel y coordinate, unsigned 16bit int
p: polarity value, boolean. False = off event, True = on event
ts: timestamp in microseconds, unsigned 64bit int
width: The width of the frame.
height: The height of the frame.
"""
with open(filename, 'rb') as f:
# Strip header
header_line = f.readline()
while header_line[0] == '#':
header_line = f.readline()
raw_data = np.fromfile(f, dtype=np.uint8)
# file already closed since using 'with' statement
total_events = len(raw_data)
full_x = np.zeros(total_events)
full_y = np.zeros(total_events)
full_p = np.zeros(total_events)
full_ts = np.zeros(total_events)
full_f = np.zeros(total_events)
TD_indices = np.zeros(total_events, dtype=np.bool_)
total_events = 0
buffer_location = 0
start_evt_ind = 0
while buffer_location < len(raw_data):
num_events = ((raw_data[buffer_location + 3].astype(np.uint32) << 24) + (raw_data[buffer_location + 2].astype(np.uint32) << 16) + (raw_data[buffer_location + 1].astype(np.uint32) << 8) + raw_data[buffer_location])
buffer_location = buffer_location + 4
start_time = ((raw_data[buffer_location + 3].astype(np.uint32) << 24) + (raw_data[buffer_location + 2].astype(np.uint32) << 16) + (raw_data[buffer_location + 1].astype(np.uint32) << 8) + raw_data[buffer_location])
buffer_location = buffer_location + 8
# Note renaming (since original is a Python built-in):
evt_type = raw_data[buffer_location:(buffer_location + 8 * num_events):8]
evt_subtype = raw_data[(buffer_location + 1):(buffer_location + 8 * num_events + 1):8]
y = raw_data[(buffer_location + 2):(buffer_location + 8 * num_events + 2):8]
x = ((raw_data[(buffer_location + 5):(buffer_location + 8 * num_events + 5):8].astype(np.uint16) << 8) + (raw_data[(buffer_location + 4):(buffer_location + 8 * num_events + 4):8]))
ts = ((raw_data[(buffer_location + 7):(buffer_location + 8 * num_events + 7):8].astype(np.uint32) << 8) + (raw_data[(buffer_location + 6):(buffer_location + 8 * num_events + 6):8]))
buffer_location = buffer_location + num_events * 8
ts = ts + start_time
overflows = np.where(evt_type == 2)
for i in range(0, len(overflows[0])):
overflow_loc = overflows[0][i]
ts[overflow_loc:] = ts[overflow_loc:] + 65536
locations = np.where((evt_type == 0) | (evt_type == 3))
TD_indices[start_evt_ind:(start_evt_ind + num_events)][locations] = True
full_x[start_evt_ind:(start_evt_ind + num_events)] = x
full_y[start_evt_ind:(start_evt_ind + num_events)] = y
full_p[start_evt_ind:(start_evt_ind + num_events)] = evt_subtype
full_ts[start_evt_ind:(start_evt_ind + num_events)] = ts
full_f[start_evt_ind:(start_evt_ind + num_events)] = evt_type
start_evt_ind = start_evt_ind + num_events
TD = Events(len(full_x[TD_indices]))
# If intefacing with Matlab, 1 must be added to x and y indices.
# due to Matlab's index convention.
TD.data.x = full_x[TD_indices] # + 1
TD.data.y = full_y[TD_indices] # + 1
TD.data.ts = full_ts[TD_indices]
TD.data.p = full_p[TD_indices]
# It appears that the polarity needs to be flipped (when results are compared with Matlab output).
# Change the polarity: 0 events become 1 events and vice versa.
TD.data.p = np.abs(TD.data.p - 1)
return TD
def main():
"""Example usage of eventvision"""
#read in some data
td, em = read_aer('0001.val')
#td = read_dataset('trainReduced/0/00002.bin')
#show the TD events
td.show_td()
#extract a region of interest...
td.data = td.extract_roi([75, 75], [50, 50], True)
#td.data = td.extract_roi([3, 3], [28, 28])
#implement a refractory period...
td.data = td.apply_refraction(0.03)
#perform some noise filtering...
td.data = td.filter_td(0.03)
#show the resulting data
td.show_td()
#write the filtered data in a format jAER can understand
td.write_j_aer('jAERdata.aedat')
#show the grayscale data
em.show_em()
#perform camera calibration
#first show the calibration pattern on the screen and make some recordings:
num_squares = 10
square_size_mm = present_checkerboard(num_squares)
#state where the recordings are what format they are in
image_directory = 'path_to_calibration_images'
image_format = '.bmp'
#using a scale is useful for visualization
scale = 4
#call the calibration function and follow the instructions provided
ret, mtx, dist, rvecs, tvecs = auto_calibrate(num_squares, square_size_mm, scale, image_directory, image_format)
if __name__ == "__main__":
main()
print 'Event-based vision module imported'