-
Notifications
You must be signed in to change notification settings - Fork 0
/
pipeline.py
440 lines (354 loc) · 17.5 KB
/
pipeline.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
#_______________________________________________________________________________
# pipeline.py 80->|
import cv2
import glob
import pickle
import numpy as np
from moviepy.editor import VideoFileClip
from PIL import ImageFont, ImageDraw, Image
#_______________________________________________________________________________
# Circular queuing class made with Numpy
class NQ():
def __init__(self,qlen,shape,dtype=np.uint8):
self.data= np.empty(shape=shape,dtype=dtype)
self.qlen= qlen
def put(self,a):
self.data= np.append(self.data,a, axis=0)
if len(self.data)>self.qlen: self.drop()
def peek(self):
return self.data[len(self.data)-1]
def drop(self):
self.data= np.delete(self.data, 0, axis=0)
def getAvg(self):
newList= np.copy(self.data[0])
newList[:,1]= np.sum(self.data[:,:,1], axis=0)/float(len(self.data))
return newList
#_______________________________________________________________________________
# Control class
class Executive():
frameCount= 0
diagScreen= None
def __init__(self, video=False, diag=False):
self.diags= diag
self.video= video
self.debugger= open('output_images/trace.txt', 'w')
self.debugger.write('Debug logger\n')
def reset(self):
self.frameCount= 0
self.debugger.close()
#_______________________________________________________________________________
# Line management class
class Line():
def __init__(self, qlen=10, yoffset=209):
self.qlen= qlen # queue length
line= []
for i in range(12): # Each line has 12 points
line.append([i*114+14, yoffset]) # Each point has an x,y pair
self.pointsQ= NQ(qlen, (0,12,2)) # Points queue
self.pointsQ.put([ line ])
self.fit= [0,0,yoffset] # Init with a straight line
# Return a y value as a function of x based on average fit coefficients
def lineF(self,x):
fit= self.fit
return fit[0]*x**2 + fit[1]*x + fit[2]
def curveF(self,x): # return radius of curvature at point x
fit= self.fit
return ((1 + (2*fit[0]*x + fit[1])**2)**1.5) / np.absolute(2*fit[0])
def getAvgPoints(self): # Return an array of average points on a line
return self.pointsQ.getAvg()
def getPrevPoints(self): # Return an array of previously enqueued points
return self.pointsQ.peek()
def putPoints(self, line):
self.pointsQ.put([ line ])
def fitToPoly(self): # Fit pointsQ to a polynomial and save coefficients
avgPoints= self.pointsQ.getAvg()
self.fit= np.polyfit(avgPoints[:,0],avgPoints[:,1], 2)
#_______________________________________________________________________________
# The Model class creates a left, center, and right line
class Model():
curRadius= 0
camY= 275
def __init__(self, width=160):
self.leftLine= Line(3, self.camY-width//2)
self.centerLine= Line(15, self.camY)
self.rightLine= Line(3, self.camY+width//2)
self.widthQs= [] # Width queues for each right-side patch
for i in range(12): # longer queue on the widths to hold through bridges
self.widthQs.append(NQ(4, (0,1), np.uint16))
# initialize with the starting width
self.widthQs[i].put([ [width] ])
#_______________________________________________________________________________
# Filter on range of white
def filterW(img, thresh=200):
lower= np.array([thresh, thresh, thresh], dtype="uint8")
upper= np.array([255, 255, 255], dtype="uint8")
mask= cv2.inRange(img, lower, upper)
return mask
#_______________________________________________________________________________
# Filter on range of yellow
def filterY(img, thresh=146):
lower= np.array([thresh, thresh, 0], dtype="uint8")
upper= np.array([255, 209, 130], dtype="uint8")
mask= cv2.inRange(img, lower, upper)
return mask
#_______________________________________________________________________________
# Convert to HLS color space and separate the S channel and filter by thresholds
def thresholdPatch(img, primary, idx, s_thresh=(200, 255)):
if primary:
hlsImg= cv2.cvtColor(img, cv2.COLOR_RGB2HLS).astype(np.float)
S_channel= hlsImg[:,:,2]
_,s_binary= cv2.threshold(S_channel.astype('uint8'), s_thresh[0], s_thresh[1], cv2.THRESH_BINARY)
# RULE: if nothing found and idx>4, then use filterY
if np.sum(s_binary)==0 and idx>4:
s_binary= filterY(img,110)
else:
s_binary= filterW(img,s_thresh[0])
# RULE: if nothing found and idx>2, then try lower threshold
if np.sum(s_binary)==0 and idx>2:
s_binary= filterW(img,s_thresh[0]-20)
return s_binary
#_______________________________________________________________________________
# Transform from perspective view to overhead view and vice versa
src= np.float32([[557,460],[557+166,460],[84+1112,670],[84,670]])
xoffset, yoffset= 25, 156
width, height= 1248, 237
dst= np.float32([[xoffset+width,yoffset+0],[xoffset+width,yoffset+height],[xoffset,yoffset+height],[xoffset,yoffset+0]])
def makeOverhead(img):
global src, dst
img_size= (img.shape[1], img.shape[0])
M= cv2.getPerspectiveTransform(src, dst)
return cv2.warpPerspective(img, M, img_size, flags=cv2.INTER_LINEAR) #INTER_NEAREST
def makePerspective(img):
global src, dst
img_size= (img.shape[1], img.shape[0])
M= cv2.getPerspectiveTransform(dst, src)
return cv2.warpPerspective(img, M, img_size, flags=cv2.INTER_LINEAR)
#_______________________________________________________________________________
# Display filter patches into diagnostics window
def diagPatch(primary,patch,idx):
global executive
if primary:
if np.sum(patch)==0:
miss= np.ones_like(patch).astype(np.uint8)
executive.diagScreen[77-idx*7:77-idx*7+7,0:58,:]= np.rot90(np.dstack(( miss*63, miss*0, miss*0)))
else:
executive.diagScreen[77-idx*7:77-idx*7+7,0:58,:]= np.rot90(np.dstack(( patch, patch, patch)))
else:
if np.sum(patch)==0:
miss= np.ones_like(patch).astype(np.uint8)
executive.diagScreen[77-idx*7:77-idx*7+7,58:58*2,:]= np.rot90(np.dstack(( miss*63, miss*0, miss*0)))
else:
executive.diagScreen[77-idx*7:77-idx*7+7,58:58*2,:]= np.rot90(np.dstack(( patch, patch, patch)))
#_______________________________________________________________________________
# Scan patch for signs of a line and return a potential y and confidence ranking
def processFilter(primary,patch,idx):
if primary:
patchT= thresholdPatch(patch,primary,idx,(100,255)) #threshold patch
else:
patchT= thresholdPatch(patch,primary,idx,(180,255)) # threshold patch for white values
diagPatch(primary,patchT,idx)
histogram= np.convolve(np.ones(5).astype(np.uint16),np.sum(patchT/7, axis=1),mode='same')
maybeY= np.round(np.argmax(histogram))
#Maximum confidence is 1275= 255*5 (a stack of 5 white pixels)
return histogram[maybeY]/1275, maybeY
#_______________________________________________________________________________
# Locate and update leftLine and rightLine in overhead xy space
def locateLines(img):
global model,executive
# generate diag screen to display patch diagnostics
executive.diagScreen= np.zeros_like(img).astype(np.uint8)
# Extract average leftPoints from leftLine
leftPoints= model.leftLine.getPrevPoints()
for idx,i in enumerate(leftPoints):
if executive.frameCount>5: # RULE: filter extreme curvatures
if idx>0 and abs(i[1]-leftPoints[idx-1][1])>7:
i[1]= leftPoints[idx-1][1]
conf,y= processFilter(True,img[i[1]-29:i[1]+29, i[0]:i[0]+7],idx)
if conf>.6:
yCandidate= i[1]-29+y
if executive.frameCount>25:
if abs(yCandidate-i[1])<10: # RULE: filter outliers
i[1]= yCandidate
else:
i[1]= yCandidate
model.leftLine.putPoints(leftPoints)
# Extract rightPoints offset by dist to center from leftPoints
rightPoints= np.copy(leftPoints)
rightPoints[:,1]= leftPoints[:,1]+(model.centerLine.getAvgPoints()[:,1]-leftPoints[:,1])*2
for idx,i in enumerate(rightPoints):
if executive.frameCount>5: # RULE: filter extreme curvatures
if idx>0 and abs(i[1]-rightPoints[idx-1][1])>7:
i[1]= rightPoints[idx-1][1]
conf,y= processFilter(False,img[i[1]-29:i[1]+29, i[0]:i[0]+7],idx)
if conf>.6:
yCandidate= i[1]-29+y
if executive.frameCount>25:
if abs(yCandidate-i[1])<15+int(round(14*idx/11)): # RULE: filter outliers
i[1]= yCandidate
model.widthQs[idx].put([ [int(round(i[1]-leftPoints[idx][1]))] ])
else:
i[1]= yCandidate
else: # RULE: right lane is average distance from left lane
avgPrevLaneWidth= np.sum(model.widthQs[idx].data)/float(len(model.widthQs[idx].data))
i[1]= leftPoints[idx][1]+avgPrevLaneWidth
model.rightLine.putPoints(rightPoints)
# Adjust center line
centerPoints= model.centerLine.getAvgPoints()
comb= np.add(rightPoints[:,1],leftPoints[:,1])
centerPoints[:,1]= np.round(comb/2).astype(np.int) #update y values
model.centerLine.putPoints(centerPoints)
#_______________________________________________________________________________
# Draw in overhead space, the found lane delimited by the lines
def drawLane(img):
global executive,model
base= np.zeros_like(img).astype(np.uint8) # generate image on which to draw
if executive.video:
laneColor= (0,0,255)
centerColor= (0,150,255)
else:
laneColor= (255,0,0)
centerColor= (255,150,0)
# Fit left line to a 2nd order polynomial
model.leftLine.fitToPoly()
leftPoints= np.copy(model.leftLine.pointsQ.data[0]) # Copy the first element to use x values
for i in leftPoints:
i[1]= model.leftLine.lineF(i[0])+6
# Fit center line to a 2nd order polynomial
model.centerLine.fitToPoly()
avgCenterO= np.copy(model.centerLine.pointsQ.data[0]) # Copy the first element to use x values
for i in avgCenterO:
i[1]= model.centerLine.lineF(i[0])-2 # offset to the left of center by 2 pixels
# Fit right line to a 2nd order polynomial
model.rightLine.fitToPoly()
rightPoints= np.copy(model.rightLine.pointsQ.data[0]) # Copy the first element to use x values
for i in rightPoints:
i[1]= model.rightLine.lineF(i[0])-6
if executive.diags: # for diagnostic view, show center line
pointStack= np.hstack((np.array([leftPoints]), np.array([ np.flipud( avgCenterO )]) ))
cv2.fillPoly(base, np.int_([pointStack]), laneColor) # draw the lane onto the overhead blank image
avgCenterO[:,1]= avgCenterO[:,1]+4 # offset to the right of center
pointStack= np.hstack((np.array([avgCenterO]), np.array([ np.flipud( rightPoints )]) ))
cv2.fillPoly(base, np.int_([pointStack]), laneColor) # draw the lane onto the overhead blank image
else:
pointStack= np.hstack((np.array([leftPoints]), np.array([ np.flipud( rightPoints )]) ))
cv2.fillPoly(base, np.int_([pointStack]), laneColor) # draw the lane onto the overhead blank image
# paint the line points
if executive.diags:
if executive.video:
spotColor= (0,150,255)
else:
spotColor= (255,150,0)
leftPoints= model.leftLine.getPrevPoints()
rightPoints= model.rightLine.getPrevPoints()
for i in leftPoints:
cv2.rectangle(base, tuple([i[0]-2,i[1]-29]), tuple([i[0]+16,i[1]+29]), spotColor, -1) #CV_FILLED
for i in rightPoints:
cv2.rectangle(base, tuple([i[0]-2,i[1]-29]), tuple([i[0]+16,i[1]+29]), spotColor, -1)
overlay= makePerspective(base) # warp the overhead to perspective space
result= cv2.addWeighted(img, 1, overlay, 0.5, 0) # annotate the original
return result
#_______________________________________________________________________________
# Compute radius using OpenCV
def radius(x1,y1, x2,y2, x3,y3):
c,r=cv2.minEnclosingCircle(np.array([[x1,y1], [x2,y2], [x3,y3]], dtype=np.float32))
return c[1]<0,r # also return true if center is to the left
#_______________________________________________________________________________
# Add computed data to the annotated screen using PIL
def drawData(img):
global executive,model
radiusMeters= 0
# Draw semi-transparent base on which to place text
pImg= Image.fromarray(img).convert('RGBA')
rImg= Image.new('RGBA', pImg.size, (0,0,0,0))
draw= ImageDraw.Draw(rImg) # get a drawing context
font= ImageFont.truetype('OpenSans-Regular.ttf', 35)
fullWhite= (255,255,255,255)
if executive.video:
attrColor= (100,195,36,255)
else:
attrColor= (36,195,100,255)
x,y= 32,661
if executive.diags:
draw.rectangle([0,0, 58*2,7*12],fill=(32,32,32,127))
draw.rectangle([ 0,630, 316,720],fill=(32,32,32,232))
draw.rectangle([321,630, 637,720],fill=(32,32,32,232))
draw.rectangle([642,630, 958,720],fill=(32,32,32,232))
draw.rectangle([964,630,1280,720],fill=(32,32,32,232))
draw.polygon([x+0,y+26.311, x+7.594,y+13.156, x+0,y+0, x+8.064,y+0, x+15.658,y+13.156, x+8.064,y+26.311], fill=attrColor)
draw.text((70,648), 'width', font=font,fill=attrColor)
avgPrevLaneWidth= np.sum(model.widthQs[0].data)/float(len(model.widthQs[0].data))
draw.text((184,648), str(int(round(avgPrevLaneWidth*3.6576/1.6))/100), font=font,fill=fullWhite)
draw.text(( 343,648), 'drift', font=font,fill=attrColor)
draw.text(( 659+11,648), 'radius', font=font,fill=attrColor)
draw.text(( 990,648), 'frame', font=font,fill=attrColor)
draw.text((1104,648), str(executive.frameCount), font=font,fill=fullWhite)
# Compute curvature radius along the polynomial (48' behind to 48' ahead)
if executive.frameCount%20==0:
left,radiusPx= radius(-640,model.centerLine.lineF(-640), 0,model.centerLine.lineF(0), 640,model.centerLine.lineF(640))
radiusMeters= int(round(radiusPx*3.6576/160))
model.curRadius= radiusMeters
if left:
model.curRadius= -model.curRadius
if radiusMeters>10000 or model.curRadius==0: # Straight section found
model.curRadius= 0
draw.text((781+11,648), '9999m', font=font,fill=fullWhite)
else:
draw.text((781+11,648), str(model.curRadius)+'m', font=font,fill=fullWhite)
# Compute and display lane drift in centimeters
centerCM= int(round((model.camY-model.centerLine.lineF(0))*36.576/16))
draw.text((423,648), str(centerCM)+'cm', font=font,fill=fullWhite)
del draw # free up the context
pImg= Image.alpha_composite(pImg, rImg)
annotated= np.array(pImg.convert('RGB'))
if executive.diags:
annotated= cv2.addWeighted(annotated, 1, executive.diagScreen, 1, 0)
return annotated
#_______________________________________________________________________________
# Additional filters for future work on challenge videos
def expandUpper(img,boost=1):
img= img.astype(np.int16) -127
img= img.clip(0)
imgNew= ((img*2)*boost).clip(0,255)
return imgNew.astype(np.uint8)
def sobelY(img, threshL=20, threshU=100):
sobely= cv2.Sobel(img, cv2.CV_16S, 0, 1) # run the derivative in y
abs_sobely= np.absolute(sobely) # absolute y derivative to accentuate horizontal lines
scaled_sobel= np.uint8(255*abs_sobely/np.max(abs_sobely))
binary= np.zeros_like(img)
binary[(scaled_sobel >threshL) & (scaled_sobel <= threshU)]= 255 # threshold y gradient
return binary
def prefilterImg(img):
imgE= expandUpper(img, boost=1)
imgGray= cv2.cvtColor(imgE,cv2.COLOR_RGB2HLS)[:,:,1]
imgB= sobelY(imgGray,20,68)
return np.dstack(( imgB,imgB,imgB ))
#_______________________________________________________________________________
# The processing pipeline to analyze and annotate an image
def process(img):
global executive,mtx,dist
# Undistort and convert to overhead space
output= cv2.undistort(img, mtx, dist, None, mtx)
warped= makeOverhead(output)
locateLines(warped)
withLane= drawLane(output)
annotated= drawData(withLane)
executive.frameCount += 1
return annotated
#_______________________________________________________________________________
# Process a video stream
def procVideo(fileName):
clip= VideoFileClip(fileName)
imgName= fileName.split('/')[1]
project_video_output= 'output_images/'+imgName
print('Processing video...')
project_video_clip= clip.fl_image(process)
project_video_clip.write_videofile(project_video_output, audio=False)
#_______________________________________________________________________________
# Executive
calibration_data= pickle.load(open("dist_pickle.p", "rb"))
mtx= calibration_data["mtx"]
dist= calibration_data["dist"]
executive= Executive(video=True, diag=True)
model= Model(width=160)
procVideo('video/project_video.mp4')
executive.reset()