-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathpipeline_phase1.py
448 lines (318 loc) · 19.1 KB
/
pipeline_phase1.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
import cv2
import numpy as np
import salama_preprocessing_techniques as prep
import salama_lane_detection_algorithm as lane
NUM_OF_PREPROCESSES_USED = None
def __pipeline_preprocesses(frame_BGR):
frame_GRAY = cv2.cvtColor(frame_BGR, cv2.COLOR_BGR2GRAY)
frame_HLS_S = cv2.cvtColor(frame_BGR, cv2.COLOR_BGR2HLS)[:, :, 2]
frame_HLS_L = cv2.cvtColor(frame_BGR, cv2.COLOR_BGR2HLS)[:, :, 1]
frame_HSV_V = cv2.cvtColor(frame_BGR, cv2.COLOR_BGR2HLS)[:, :, 1]
frame_LS_hybrid = cv2.addWeighted(frame_HLS_L, 1.5, frame_HLS_S, 2.5, -255)
frame_VS_hybrid = cv2.addWeighted(frame_HSV_V, 1.5, frame_HLS_S, 2.5, -255)
preprocessed_images = []
preprocessed_images.append(prep.color_thresholded_edges_pre(frame_BGR, prep.HLS_COLORSPACE_THRESH)[0])
preprocessed_images.append(prep.color_thresholded_edges_pre(frame_BGR, prep.LAB_COLORSPACE_THRESH)[0])
# preprocessed_images.append(prep.adaptive_thresholding_pre(frame_HLS_S))
preprocessed_images.append(prep.adaptive_thresholding_pre(frame_LS_hybrid))
preprocessed_images.append(prep.adaptive_thresholding_pre(frame_VS_hybrid))
preprocessed_images.append(prep.sobel_pre(frame_HLS_S, (1, 3))[0])
preprocessed_images.append(prep.sobel_pre(frame_LS_hybrid, (1, 5))[0])
preprocessed_images.append(prep.sobel_pre(frame_VS_hybrid, (1, 5))[0])
preprocessed_images.append(prep.morph_top_hat_pre(frame_BGR)[0])
preprocessed_images.append(prep.morph_inv_black_hat_pre(frame_BGR)[0])
preprocessed_images.append(prep.canny_raw_pre(frame_GRAY))
global NUM_OF_PREPROCESSES_USED
NUM_OF_PREPROCESSES_USED = len(preprocessed_images)
return preprocessed_images
# def __pipeline_score(previous_lane_data: lane.lane_data, next_lane_data: lane.lane_data):
# union = 0
# intersection = 0
# for y in range(0, next_lane_data.lane_height):
# for x in range(0, next_lane_data.lane_width):
# match_next = (
# x > next_lane_data.left_lane_polyfit_pts[y]
# and
# x < next_lane_data.right_lane_polyfit_pts[y]
# )
# match_prev = (
# x > previous_lane_data.left_lane_polyfit_pts[y]
# and
# x < previous_lane_data.right_lane_polyfit_pts[y]
# )
# union += (match_next or match_prev)
# intersection += (match_next and match_prev)
# return ((intersection / union) if union != 0 and intersection != 0 else inf, next_lane_data)
##################################################################################################
# def __pipeline_score(previous_lane_data: lane.lane_data, next_lane_data: lane.lane_data):
# score = [0, 0]
# for y in range(0, next_lane_data.lane_height):
# score[0] += abs(previous_lane_data.left_lane_polyfit_pts[y] - next_lane_data.left_lane_polyfit_pts[y])
# score[1] += abs(previous_lane_data.right_lane_polyfit_pts[y] - next_lane_data.right_lane_polyfit_pts[y])
# return score
##################################################################################################
def __pipeline_get_lanes_scores(previous_lane_data: lane.lane_data,
next_lane_candidates: list[cv2.Mat],
left_lane_mask, right_lane_mask):
candidates_scored = []
for candidate in next_lane_candidates:
# FIXME: Hough is too harsh, but I need its guidance nevertheless
_, candidate_edges_canvas = prep.hough_transform_raw_pre(candidate)
# candidate_edges_canvas = candidate
candidate_edges_canvas_left_lane = cv2.bitwise_and(candidate_edges_canvas, left_lane_mask)
candidate_edges_canvas_left_lane_pts = np.argwhere(candidate_edges_canvas_left_lane != 0)
# if candidate_edges_canvas_left_lane_pts.size == 0: continue
candidate_edges_canvas_right_lane = cv2.bitwise_and(candidate_edges_canvas, right_lane_mask)
candidate_edges_canvas_right_lane_pts = np.argwhere(candidate_edges_canvas_right_lane != 0)
# if candidate_edges_canvas_right_lane_pts.size == 0: continue
if candidate_edges_canvas_left_lane_pts.size == 0 or candidate_edges_canvas_right_lane_pts.size == 0:
candidates_scored.append([None, None, np.full_like(candidate_edges_canvas, 63)])
continue
candidate_lane_data = lane.lane_data(
lane_height=previous_lane_data.lane_height,
lane_width=previous_lane_data.lane_width,
l_pts=candidate_edges_canvas_left_lane_pts,
r_pts=candidate_edges_canvas_right_lane_pts,
lane_polyfit_rank=1
)
# FIXME: This scoring method is too sensitive to the shaky camera
# Since the algorithm depends on the difference in running scores
# This sensitivity messes everything up
# score = [0, 0]
# for y in range(0, previous_lane_data.lane_height):
# score[0] += abs(previous_lane_data.left_lane_polyfit_pts[y] -
# candidate_lane_data.left_lane_polyfit_pts[y])
# score[1] += abs(previous_lane_data.right_lane_polyfit_pts[y] -
# candidate_lane_data.right_lane_polyfit_pts[y])
################################################################################
approx_error_left = 0
left_lane_coeffs = list(zip(previous_lane_data.left_lane_polyfit_coeffs,
candidate_lane_data.left_lane_polyfit_coeffs))
for previous_coeff, candidate_coeff in left_lane_coeffs:
approx_error_left += 10000 * abs((previous_coeff - candidate_coeff) / previous_coeff)
approx_error_right = 0
right_lane_coeffs = list(zip(previous_lane_data.right_lane_polyfit_coeffs,
candidate_lane_data.right_lane_polyfit_coeffs))
for previous_coeff, candidate_coeff in right_lane_coeffs:
approx_error_right += 10000 * abs((previous_coeff - candidate_coeff) / previous_coeff)
score = [approx_error_left, approx_error_right]
candidates_scored.append([score, candidate_lane_data, candidate_edges_canvas])
return candidates_scored
def get_next_lane(previous_lane_data: lane.lane_data, frame_BGR: cv2.Mat, skycutoff, hoodcutoff, debugging):
try:
left_ROI = previous_lane_data.extract_left_ROI(ROI_width=min(
np.shape(frame_BGR)[1] // 32 + get_next_lane.previous_lanes_average_score_left // 160,
np.shape(frame_BGR)[1] // 16)
)
# left_ROI = previous_lane_data.extract_left_ROI(ROI_width=max(
# np.shape(frame_BGR)[1] // 32 - get_next_lane.previous_lanes_average_score_left // 160,
# np.shape(frame_BGR)[1] // 16)
# )
right_ROI = previous_lane_data.extract_right_ROI(ROI_width=max(
np.shape(frame_BGR)[1] // 16 - get_next_lane.previous_lanes_average_score_right // 160,
np.shape(frame_BGR)[1] // 32)
)
except AttributeError:
left_ROI = previous_lane_data.extract_left_ROI(ROI_width=np.shape(frame_BGR)[1] // 16)
right_ROI = previous_lane_data.extract_right_ROI(ROI_width=np.shape(frame_BGR)[1] // 16)
left_lane_mask = previous_lane_data.extract_mask(left_ROI)
right_lane_mask = previous_lane_data.extract_mask(right_ROI)
# FIXME: The perpendiculars implementation messed this up
# left_lane_mask_topleft_area = cv2.fillPoly(np.zeros_like(left_lane_mask),
# pts=[np.array([[0, 0],
# left_ROI[0],
# left_ROI[3]])],
# color=255)
# right_lane_mask_topright_area = cv2.fillPoly(np.zeros_like(right_lane_mask),
# pts=[np.array([[np.shape(right_lane_mask)[1], 0],
# right_ROI[1],
# right_ROI[2]])],
# color=255)
# lane_masks_intersection = cv2.bitwise_and(left_lane_mask, right_lane_mask)
# left_lane_mask = cv2.subtract(left_lane_mask, right_lane_mask_topright_area)
# left_lane_mask = cv2.subtract(left_lane_mask, lane_masks_intersection)
# right_lane_mask = cv2.subtract(right_lane_mask, left_lane_mask_topleft_area)
# right_lane_mask = cv2.subtract(right_lane_mask, lane_masks_intersection)
y_cutoff = 0
for y in range(previous_lane_data.lane_height - 1, -1, -1):
if cv2.countNonZero(cv2.bitwise_and(left_lane_mask[y], right_lane_mask[y])) > 1:
y_cutoff = y
break
for y in range(y_cutoff, -1, -1):
left_lane_mask[y] = 0
right_lane_mask[y] = 0
next_lane_candidates = __pipeline_preprocesses(frame_BGR[skycutoff:hoodcutoff])
scored_candidates = __pipeline_get_lanes_scores(previous_lane_data, next_lane_candidates,
left_lane_mask, right_lane_mask)
best_candidate_left = min(filter(lambda _: _[0] != None, scored_candidates), key=lambda c: c[0][0])
best_candidate_right = min(filter(lambda _: _[0] != None, scored_candidates), key=lambda c: c[0][1])
get_next_lane.previous_lanes_average_score_left = 0
get_next_lane.previous_lanes_average_score_right = 0
for candidate in scored_candidates:
if candidate[0] != None:
get_next_lane.previous_lanes_average_score_left += candidate[0][0]
get_next_lane.previous_lanes_average_score_right += candidate[0][1]
get_next_lane.previous_lanes_average_score_left /= len(scored_candidates)
get_next_lane.previous_lanes_average_score_right /= len(scored_candidates)
try:
get_next_lane.best_candidate_score_left_memory = get_next_lane.best_candidate_score_left_memory[1:]
get_next_lane.best_candidate_score_left_memory.append(best_candidate_left[0][0])
get_next_lane.best_candidate_score_right_memory = get_next_lane.best_candidate_score_right_memory[1:]
get_next_lane.best_candidate_score_right_memory.append(best_candidate_right[0][1])
except AttributeError:
get_next_lane.best_candidate_score_left_memory = [1000] * 5
get_next_lane.best_candidate_score_right_memory = [1000] * 5
# TODO: If best score is below some threshold (1.25x w.r.t memory average sounds good)...
# ...go ahead with ensemble, in increasing voting threshold order
# If something gets a better score (that is below a slightly larger threshold), take it obviously lol
# However if not, just re-use previous lane data
use_previous = False
voting_ensemble_candidates = []
if best_candidate_left[0][0] > 1.25 * np.mean(get_next_lane.best_candidate_score_left_memory)\
or best_candidate_right[0][1] > 1.25 * np.mean(get_next_lane.best_candidate_score_right_memory):
for i in range(1, len(next_lane_candidates) + 1):
voting_ensemble_candidates.append(prep.edge_voting_ensemble(next_lane_candidates, i))
# Reminder that __pipeline_lanes_score may or may not apply Hough Transform based on my mood hehe
scored_voting_ensemble_candidates = __pipeline_get_lanes_scores(previous_lane_data, voting_ensemble_candidates,
left_lane_mask, right_lane_mask)
best_voting_ensemble_candidate_left =\
min(filter(lambda _: _[0] != None, scored_voting_ensemble_candidates),
key=lambda c: c[0][0])
best_voting_ensemble_candidate_right =\
min(filter(lambda _: _[0] != None, scored_voting_ensemble_candidates),
key=lambda c: c[0][1])
best_candidate_left = min(best_candidate_left, best_voting_ensemble_candidate_left, key=lambda c: c[0][0])
best_candidate_right = min(best_candidate_right, best_voting_ensemble_candidate_right, key=lambda c: c[0][1])
if best_candidate_left[0][0] > 1.10 * np.mean(get_next_lane.best_candidate_score_left_memory) or \
best_candidate_right[0][1] > 1.10 * np.mean(get_next_lane.best_candidate_score_right_memory):
use_previous = True
pass
next_lane_data = previous_lane_data if use_previous else lane.lane_data(
lane_height=best_candidate_left[1].lane_height, # arbitrary choice, either of them work
lane_width=best_candidate_left[1].lane_width, # same ^
l_pts=best_candidate_left[1].left_lane_pts,
r_pts=best_candidate_right[1].right_lane_pts,
lane_polyfit_rank=1
)
next_lane_canvas = next_lane_data.draw(frame_BGR, skycutoff, hoodcutoff, debugging)
if debugging:
all_candidates_scored = [scored_candidates]
if len(voting_ensemble_candidates) != 0: all_candidates_scored.append(scored_voting_ensemble_candidates)
all_candidates_marked_RGB = []
for scored_candidates in all_candidates_scored:
candidates_marked_RGB = []
for candidate in scored_candidates:
if candidate[0] == None:
candidate_RGB = cv2.cvtColor(candidate[2], cv2.COLOR_GRAY2RGB)
else:
best_left = (candidate[1] == best_candidate_left[1])
best_right = (candidate[1] == best_candidate_right[1])
candidate_left = cv2.bitwise_and(candidate[2], left_lane_mask)
candidate_right = cv2.bitwise_and(candidate[2], right_lane_mask)
candidate_left_RGB = cv2.cvtColor(candidate_left, cv2.COLOR_GRAY2RGB)
candidate_right_RGB = cv2.cvtColor(candidate_right, cv2.COLOR_GRAY2RGB)
left_lane_mask_RGB = cv2.cvtColor(left_lane_mask, cv2.COLOR_GRAY2RGB)
right_lane_mask_RGB = cv2.cvtColor(right_lane_mask, cv2.COLOR_GRAY2RGB)
if best_left:
candidate_left_RGB[:, :, 0] = 0
candidate_left_RGB[:, :, 1] = 0
left_lane_mask_RGB[:, :, 0] = 0
left_lane_mask_RGB[:, :, 1] = 0
if best_right:
candidate_right_RGB[:, :, 1] = 0
candidate_right_RGB[:, :, 2] = 0
right_lane_mask_RGB[:, :, 1] = 0
right_lane_mask_RGB[:, :, 2] = 0
candidate_RGB = cv2.bitwise_or(candidate_left_RGB, candidate_right_RGB)
lane_masks_RGB = cv2.bitwise_or(left_lane_mask_RGB, right_lane_mask_RGB)
candidate_RGB = cv2.addWeighted(candidate_RGB, 1, lane_masks_RGB, 0.15, 8)
candidates_marked_RGB.append(candidate_RGB)
all_candidates_marked_RGB.append(candidates_marked_RGB)
# Left score
frame_BGR = cv2.putText(
img=frame_BGR,
text=f'B: {str(int(best_candidate_left[0][0])).zfill(5)}, '
+ f'A: {str(int(np.mean(get_next_lane.previous_lanes_average_score_left))).zfill(5)}, '
+ f'M(B): {str(int(np.mean(get_next_lane.best_candidate_score_left_memory))).zfill(5)}',
bottomLeftOrigin=False, # when false, it is at the top-left corner.
org=(4, 16),
fontFace=cv2.FONT_HERSHEY_PLAIN,
fontScale=1,
color=(0, 0, 255), # BGR
lineType=cv2.LINE_AA
)
# Right score
frame_BGR = cv2.putText(
img=frame_BGR,
text=f'B: {str(int(best_candidate_right[0][1])).zfill(5)}, '
+ f'A: {str(int(np.mean(get_next_lane.previous_lanes_average_score_right))).zfill(5)}, '
+ f'M(B): {str(int(np.mean(get_next_lane.best_candidate_score_right_memory))).zfill(5)}',
bottomLeftOrigin=False, # when false, it is at the top-left corner.
org=(4, 32),
fontFace=cv2.FONT_HERSHEY_PLAIN,
fontScale=1,
color=(255, 0, 0), # BGR
lineType=cv2.LINE_AA
)
candidates_preview = []
for candidates_marked_RGB in all_candidates_marked_RGB:
candidates_marked_RGB = cv2.vconcat(candidates_marked_RGB)
candidates_marked_RGB = cv2.resize(
candidates_marked_RGB,
(int(np.shape(candidates_marked_RGB)[1] * (np.shape(frame_BGR)[0] / np.shape(candidates_marked_RGB)[0])),
np.shape(frame_BGR)[0])
)
candidates_preview.append(candidates_marked_RGB)
# Preview no longer done at this stage
# cv2.imshow(
# "Debug Mode Preview",
# cv2.hconcat(
# [
# frame_BGR,
# cv2.resize(
# candidates_preview,
# (int(np.shape(candidates_preview)[1] * (np.shape(frame_BGR)[0] / np.shape(candidates_preview)[0])),
# np.shape(frame_BGR)[0])
# )
# ]
# )
# )
# cv2.waitKey(1000)
return next_lane_data, next_lane_canvas, candidates_preview if debugging else None
def get_initial_lane(first_frame_BGR):
HOODCUTOFF, _ = prep.get_hood_cutoff(first_frame_BGR)
SKYCUTOFF, _, _ = prep.get_sky_cutoff(first_frame_BGR[:HOODCUTOFF], (1, 3))
_, detection_ready_frame =\
prep.hough_transform_raw_pre(
prep.edge_voting_ensemble(
__pipeline_preprocesses(
first_frame_BGR[SKYCUTOFF:HOODCUTOFF]
),
voting_threshold=6
)
)
detected_lane_data = lane.peeking_center_detect(detection_ready_frame, 1)
return detected_lane_data, HOODCUTOFF, SKYCUTOFF
# def alt_pipeline(INPUT_PATH, OUTPUT_PATH):
# from moviepy.editor import VideoFileClip
# video_input = VideoFileClip(INPUT_PATH)
# HOODCUTOFF, _ = pre.get_hood_cutoff(video_input.get_frame(0))
# SKYCUTOFF, _, _ = pre.get_sky_cutoff(video_input.get_frame(0)[:HOODCUTOFF], (1, 3))
# def process_frame(frame_BGR):
# preprocessed_frame, _, _ = pre.color_thresholding_preprocess(frame_BGR[SKYCUTOFF:HOODCUTOFF], [
# # H, L, S
# [(18, 22), (0, 255), (15, 255)], # Yellows
# [(0, 255), (191, 255), (127, 255)], # Whites
# ])
# left_lane_polyfit_pts, right_lane_polyfit_pts, polyfit_range, \
# left_lane_pts_x, left_lane_pts_y, left_peeking_center_trace_x, left_peeking_center_trace_y, \
# right_lane_pts_x, right_lane_pts_y, right_peeking_center_trace_x, right_peeking_center_trace_y = \
# salama_lane_detection_algorithm.peeking_center_detect(preprocessed_frame)
# lane = np.zeros_like(frame_BGR[:HOODCUTOFF])
# for y in range(0, np.shape(preprocessed_frame)[0], 1):
# for x in range(int(left_lane_polyfit_pts[y]) + 1, int(right_lane_polyfit_pts[y])):
# if x > 0 and x < np.shape(frame_BGR)[1]:
# lane[SKYCUTOFF + y][x] = [0, 31 + 224 / (np.shape(preprocessed_frame)[0] - y), 0]
# return cv2.addWeighted(lane, 1, frame_BGR[:HOODCUTOFF], 1, 0)
# processed_video = video_input.fl_image(process_frame)
# processed_video.write_videofile(f"project_video_output.mp4", audio=False)