def __init__(self, imgMarginWidth=320, historyDepth=5, margin=100, windowSplit=2, winCount=9,
searchPortion=1., veHiDepth=30, pointSize=64,
groupThrd=10, groupDiff=.1, confidenceThrd=.7):
self.imgProcessor = ip.Processing()
self.warper = ip.Warping()
self.imgMarginWidth = imgMarginWidth
'''
self.lineLeft = Line(lineSpace=LineSpace.LEFT, historyDepth=historyDepth,
margin=margin, windowSplit=windowSplit, winCount=winCount,
searchPortion=searchPortion)
self.lineRight = Line(lineSpace=LineSpace.RIGHT, historyDepth=historyDepth,
margin=margin, windowSplit=windowSplit, winCount=winCount,
searchPortion=searchPortion)
'''
self.fitHistory = []
# ====| PROJECT 5 - VEHICLE DETECTION |====
# Adding Vehicle scanner
self.scanner = VehicleScanner(pointSize=pointSize,
veHiDepth=veHiDepth, groupThrd=groupThrd, groupDiff=groupDiff,
confidenceThrd=confidenceThrd)
class Detector:
def __init__(self, imgMarginWidth=320, historyDepth=5, margin=100, windowSplit=2, winCount=9,
searchPortion=1., veHiDepth=30, pointSize=64,
groupThrd=10, groupDiff=.1, confidenceThrd=.7):
self.imgProcessor = ip.Processing()
self.warper = ip.Warping()
self.imgMarginWidth = imgMarginWidth
'''
self.lineLeft = Line(lineSpace=LineSpace.LEFT, historyDepth=historyDepth,
margin=margin, windowSplit=windowSplit, winCount=winCount,
searchPortion=searchPortion)
self.lineRight = Line(lineSpace=LineSpace.RIGHT, historyDepth=historyDepth,
margin=margin, windowSplit=windowSplit, winCount=winCount,
searchPortion=searchPortion)
'''
self.fitHistory = []
# ====| PROJECT 5 - VEHICLE DETECTION |====
# Adding Vehicle scanner
self.scanner = VehicleScanner(pointSize=pointSize,
veHiDepth=veHiDepth, groupThrd=groupThrd, groupDiff=groupDiff,
confidenceThrd=confidenceThrd)
def preProcess(self, src):
"""
Creates binary with combined threshold and applies
perspective transform for bird-eye view
:param src: source color image (assuming it's been previously undistorted)
:return: bird-eye view binary
"""
imgHeight = src.shape[0]
binary = ip.Thresholding.combiThreshold(src=src)
filler = np.zeros((imgHeight, self.imgMarginWidth), dtype=np.uint8)
binary = np.hstack((filler, binary, filler))
binaryWarp = self.warper.birdEye(img=binary, leftShift=self.imgMarginWidth)
return binaryWarp
@staticmethod
def sanityCheckPass(fitL, fitR):
"""
Checking for left and right curvatures similarity and lane width to be within a given range
:param fitL: left curve polynomial parameters
:param fitR: right curve polynomial parameters
:return: True or False
"""
tolerance = 1.2
maxCurvDelta = 0.006 # 0.000666 * tolerance
laneWidthRange = (800 / tolerance, 1000 * tolerance)
imgBottom = 719
if fitR is None or fitL is None:
return False
al = fitL[0]
ar = fitR[0]
curvDelta = abs(ar - al)
if curvDelta > maxCurvDelta:
return False
else:
bl = fitL[1]
br = fitR[1]
cl = fitL[2]
cr = fitR[2]
leftBottomX = al * (imgBottom ** 2) + bl * imgBottom + cl
rightBottomX = ar * (imgBottom ** 2) + br * imgBottom + cr
laneWidth = rightBottomX - leftBottomX
return (rightBottomX > leftBottomX) & (
laneWidthRange[1] >= laneWidth >= int(laneWidthRange[0]))
def addLanePoly(self, srcShape, dst, fitLeft, fitRight):
"""
Adding lane polygon to an empty image with the shape of bird-eye view with
subsequent perspective projection to destination image (clipping previously added left and right fillers)
:param srcShape: shape of bird-eye view (may be wider than original image)
:param dst: original color (assumed previously undistorted) image
:param fitLeft: left curve polynomial parameters
:param fitRight: right curve polynomial parameters
:return: color image with projected lane
"""
lanePolyImg, leftOutstand, _ = ip.Drawing.addPolygon(srcShape=srcShape, lFit=fitLeft,
rFit=fitRight, stepCount=10, color=(0, 255, 0))
imgH = dst.shape[1]
lShift = self.imgMarginWidth + leftOutstand
perspective = self.warper.perspective(lanePolyImg,
leftShift=lShift)[:, lShift:lShift + imgH, :]
return cv2.addWeighted(dst, 1, perspective, 0.5, 0)
def addPip(self, pipImage, dstImage, pipAlpha=0.5, pipResizeRatio=0.3, origin=(20, 20)):
"""
Adding small Picture-in-picture binary bird-eye projection with search areas and found lines embedded
:param pipImage: original binary bird-eye projection with search areas and found lines embedded
:param dstImage: destination color image (assumed undistorted)
:param pipAlpha: pip alpha
:param pipResizeRatio: pip scale
:param origin: coordinates of upper-left corner of small picture
:return: color image with P-i-P embedded
"""
smallPip = self.imgProcessor.resize(src=pipImage, ratio=pipResizeRatio)
pipHeight = smallPip.shape[0]
pipWidth = smallPip.shape[1]
backGround = dstImage[origin[1]:origin[1] + pipHeight, origin[0]:origin[0] + pipWidth]
blend = np.round(backGround * (1 - pipAlpha), 0) + np.round(smallPip * pipAlpha, 0)
blend = np.minimum(blend, 255)
dstImage[origin[1]:origin[1] + pipHeight, origin[0]:origin[0] + pipWidth] = blend
# return dstImage
def addOffsetStamp(self, leftFit, rightFit, image, origin, color=(255, 255, 255), fontScale=1.0, thickness=1):
"""
Evaluating camera offset and adding it to a given image
:param thickness: line thickness
:param leftFit: left curve polynomial parameters
:param rightFit: right curve polynomial parameters
:param image: image where data being added
:param origin: upper-left corner of the offset stamp
:param color: stamp color
:param fontScale: font scale
:return: void (adds text to passed image)
"""
imgW = image.shape[1]
imgH = image.shape[0]
yBottom = imgH - 1
cameraCenter = imgW / 2
lBottomX = aux.funcSpace(argSpace=yBottom, fitParams=leftFit) - self.imgMarginWidth
rBottomX = aux.funcSpace(argSpace=yBottom, fitParams=rightFit) - self.imgMarginWidth
laneWidth = rBottomX - lBottomX
scaleX = 3.7 / laneWidth
laneCenter = (lBottomX + rBottomX) / 2
offSet = (cameraCenter - laneCenter) * scaleX
aux.putText(img=image, text='Estimated Vehicle Offset: {:.2f} m'.format(offSet),
origin=origin, color=color, scale=fontScale, thickness=thickness)
@staticmethod
def addCurvatureStamp(leftFit, rightFit, image, origin, color=(255, 255, 255), fontScale=1.0, thickness=1):
"""
Evaluating lane curvature and adding it to a given image
:param thickness: text thickness
:param leftFit: left curve polynomial parameters
:param rightFit: right curve polynomial parameters
:param image: image where data being added
:param origin: upper-left corner of the offset stamp
:param color: stamp color
:param fontScale: font scale
:return: void (adds text to passed image)
"""
imgH = image.shape[0]
yBottom = imgH - 1
scaleY = 27 / imgH # meters per pixel
leftCurvature = aux.curvature(fitParams=leftFit, variable=yBottom, scale=scaleY)
rightCurvature = aux.curvature(fitParams=rightFit, variable=yBottom, scale=scaleY)
curvature = (leftCurvature + rightCurvature) / 2
aux.putText(img=image, text='Estimated Lane Curvature: {:.1f} m'.format(round(curvature / 100, 1) * 100),
origin=origin, color=color, scale=fontScale, thickness=thickness)
def embedDetections(self, src, pipParams=None):
"""
Main 'pipeline' for adding Lane polygon AND detected vehicles to the original image
:param src: original image
:param pipParams: alpha and scale ratios
:return: undistorted color image with the Lane embedded.
"""
# 1. Undistortion
img = self.imgProcessor.undistort(src=src)
"""
1a. ====| PROJECT 5 - VEHICLE DETECTION |====
Getting vehicle boxes
"""
vBoxes, heatMap = self.scanner.relevantBoxes(src=img)
# 2. Binary (bird-eye projection)
'''
binary = self.preProcess(src=img)
# 3. Getting fits
#currFitLeft, leftFitType, leftBin = self.lineLeft.getFit(src=binary)
#currFitRight, rightFitType, rightBin = self.lineRight.getFit(src=binary)
# 4. Evaluation
sanityPass = self.sanityCheckPass(currFitLeft, currFitRight)
if sanityPass:
# Adding Lane polygon if sanity check passed
img = self.addLanePoly(srcShape=binary.shape, dst=img, fitLeft=currFitLeft, fitRight=currFitRight)
else:
# Evaluation if box-search re-scan justified without resetting history of previous fits
if self.lineLeft.reScanJustified():
currFitLeft, leftFitType, leftBin = self.lineLeft.reScanWithPrimary(src=binary)
else:
# Resetting fits and starting from scratch with box-search
self.lineLeft.resetFits()
currFitLeft, leftFitType, leftBin = self.lineLeft.getFit(src=binary)
# Same for right fit.
if self.lineRight.reScanJustified():
currFitRight, rightFitType, rightBin = self.lineRight.reScanWithPrimary(src=binary)
else:
self.lineRight.resetFits()
currFitRight, rightFitType, rightBin = self.lineRight.getFit(src=binary)
# Evaluation after second search attempt
# sanityPass, laneWidth = self.sanityCheckPass(currFitLeft, currFitRight)
sanityPass = self.sanityCheckPass(currFitLeft, currFitRight)
# Adding Lane polygon if sanity check passed. Otherwise, simply no poly added
if sanityPass:
img = self.addLanePoly(srcShape=binary.shape, dst=img, fitLeft=currFitLeft, fitRight=currFitRight)
'''
# ====| PROJECT 5 - VEHICLE DETECTION |====
# Drawing vehicle boxes
aux.drawBoxes(img=img, bBoxes=vBoxes)
# Upper left corner where starting to add pip and telemetry
origin = (20, 20)
# Adding PIP
if pipParams is not None:
alpha = pipParams['alpha']
ratio = pipParams['scaleRatio']
# Combining bins from left and right fit searches
#commonBin = cv2.addWeighted(src1=leftBin, alpha=0.5, src2=rightBin, beta=0.5, gamma=1.0)
# To keep for subsequent telemetry stamps
#pipHeight = int(commonBin.shape[0] * ratio)
heatWidth = int(heatMap.shape[1] * ratio)
# Lane Detection Picture-in-Picture
'''
self.addPip(pipImage=commonBin, dstImage=img,
pipAlpha=alpha, pipResizeRatio=ratio, origin=origin)
'''
# ====| PROJECT 5 - VEHICLE DETECTION |====
# Vehicle Detection Picture-in-Picture
self.addPip(pipImage=heatMap, dstImage=img,
pipAlpha=alpha, pipResizeRatio=ratio,
origin=(img.shape[1] - heatWidth - 20, 20))
'''
if currFitLeft is not None and currFitRight is not None:
self.addCurvatureStamp(leftFit=currFitLeft, rightFit=currFitRight,
image=img, origin=(20, pipHeight + 40), fontScale=0.66, thickness=2,
color=(0, 255, 0))
self.addOffsetStamp(leftFit=currFitLeft, rightFit=currFitRight,
image=img, origin=(20, pipHeight + 70), fontScale=0.66, thickness=2,
color=(0, 255, 0))
'''
return img