def update(self, frame, pos=None, rate=0.125):
# Crop template image from last position
(x, y), (w, h) = self.pos if pos is None else pos, self.size
self.last_img = img = cv2.getRectSubPix(frame, (w, h), (x, y))
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
img = self.preprocess(img)
# Correlate, find position of object
self.last_resp, (dx, dy), self.psr = self.correlate(img)
# Break if lost tracking (don't update filter)
self.good = self.psr > 8.0
if not self.good:
return
# Cut out new image based on tracked location
self.pos = x+dx, y+dy
self.last_img = img = cv2.getRectSubPix(frame, (w, h), self.pos)
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
# Preprocess, get DFT
img = self.preprocess(img)
A = cv2.dft(img, flags=cv2.DFT_COMPLEX_OUTPUT)
H1 = cv2.mulSpectrums(self.G, A, 0, conjB=True) # G x F*
H2 = cv2.mulSpectrums( A, A, 0, conjB=True) # F x F*
# Get weighted average based on the rate (using the new image)
self.H1 = self.H1 * (1.0-rate) + H1 * rate
self.H2 = self.H2 * (1.0-rate) + H2 * rate
# Update filter
self.update_kernel()
def splitImage(edges, blur, tag="", margin=4):
h_projection = hprojection(edges)
v_projection = vprojection(edges)
top, bottom = cropProjection(h_projection)
left, right = cropProjection(v_projection)
# plt.imshow(edges,cmap = 'gray')
# plt.plot(range(0, len(vprojection)), vprojection, 'r')
# plt.plot(hprojection, range(0, len(hprojection)), 'b')
# plt.show()
regions = splitProjection(v_projection, left, right)
print tag, left, right, top, bottom
# print regions
# print v_projection[1270:1450]
if len(tag) == 0:
return regions, left, right, top, bottom
for region in regions:
left, leftEnd = region
if (leftEnd - left) > 220 and (leftEnd - left) < 300:
width = (leftEnd - left) / 2
cr_img = cv2.getRectSubPix(
blur, (width + margin, bottom - top + margin), (width / 2 + left, (top + bottom) / 2)
)
cv2.imwrite("crop%s-%d.png" % (tag, left), cr_img)
cr_img = cv2.getRectSubPix(
blur, (width + margin, bottom - top + margin), (leftEnd - (width / 2), (top + bottom) / 2)
)
cv2.imwrite("crop%s-%d.png" % (tag, left + width), cr_img)
else:
cr_img = cv2.getRectSubPix(
blur, (leftEnd - left + margin, bottom - top + margin), ((leftEnd + left) / 2, (top + bottom) / 2)
)
cv2.imwrite("crop%s-%d.png" % (tag, left), cr_img)
return regions, left, right, top, bottom
def update(self, frame, rate=0.125, img_override=None):
(x, y), (w, h) = self.pos, self.size
if img_override is None:
self.last_img = img = cv2.getRectSubPix(frame, (w, h), (x, y))
else:
self.last_img = img = img_override
img = self.preprocess(img)
self.last_resp, (dx, dy), self.psr = self.correlate(img)
self.good = self.psr > self.MIN_PSR
if not self.good:
return
self.pos = x + dx, y + dy
self.last_img = img = cv2.getRectSubPix(frame, (w, h), self.pos)
img = self.preprocess(img)
A = cv2.dft(img, flags=cv2.DFT_COMPLEX_OUTPUT)
H1 = cv2.mulSpectrums(self.G, A, 0, conjB=True)
H2 = cv2.mulSpectrums(A, A, 0, conjB=True)
self.H1 = self.H1 * (1.0 - rate) + H1 * rate
self.H2 = self.H2 * (1.0 - rate) + H2 * rate
self.update_kernel()
def crop(self, rect):
sourceImage = self.getOriginal()
image = sourceImage.getOpenCV()
box = cv2.boxPoints(rect)
box = np.int0(box)
W = rect[1][0]
H = rect[1][1]
Xs = [i[0] for i in box]
Ys = [i[1] for i in box]
x1 = min(Xs)
x2 = max(Xs)
y1 = min(Ys)
y2 = max(Ys)
angle = rect[2]
# Center of rectangle in source image
center = ((x1 + x2) / 2, (y1 + y2) / 2)
# Size of the upright rectangle bounding the rotated rectangle
size = (x2 - x1, y2 - y1)
M = cv2.getRotationMatrix2D((size[0] / 2, size[1] / 2), angle, 1.0)
# Cropped upright rectangle
cropped = cv2.getRectSubPix(image, size, center)
cropped = cv2.warpAffine(cropped, M, size)
croppedRotated = cv2.getRectSubPix(cropped, (int(W), int(H)), (size[0] / 2, size[1] / 2))
return croppedRotated
def normCrossCorrelation(img1, img2, pt0, pt1, status, winsize, method=cv2.cv.CV_TM_CCOEFF_NORMED):
"""
**SUMMARY**
Calculates normalized cross correlation for every point.
**PARAMETERS**
img1 - Image 1.
img2 - Image 2.
pt0 - vector of points of img1
pt1 - vector of points of img2
status - Switch which point pairs should be calculated.
if status[i] == 1 => match[i] is calculated.
else match[i] = 0.0
winsize- Size of quadratic area around the point
which is compared.
method - Specifies the way how image regions are compared. see cv2.matchTemplate
**RETURNS**
match - Output: Array will contain ncc values.
0.0 if not calculated.
"""
nPts = len(pt0)
match = np.zeros(nPts)
for i in np.argwhere(status):
i = i[0]
patch1 = cv2.getRectSubPix(img1,(winsize,winsize),tuple(pt0[i]))
patch2 = cv2.getRectSubPix(img2,(winsize,winsize),tuple(pt1[i]))
match[i] = cv2.matchTemplate(patch1,patch2,method)
return match
def removeBackground(self,img,x,y,w,h,postit):
mask = np.zeros((img.shape[0], img.shape[1], 3), np.uint8)
cv2.drawContours( mask, [postit], -1, (255,255,255),-1)
img2 = cv2.getRectSubPix(img, (w, h), (x+w/2, y+h/2))
postitMask = cv2.getRectSubPix(mask, (w, h), (x+w/2, y+h/2))
img2channels = cv2.split(img2)
postitMaskChannels = cv2.split(postitMask)
img2channels.append(postitMaskChannels.pop())
img2 = cv2.merge(img2channels)
return img2
def normCrossCorrelation(self, img1, img2, points1, points2):
i = 0
chelper = CHelper()
while i < len(self.points):
if self.status[i] == 1:
rec0 = cv2.getRectSubPix(img1, (10, 10), (points1[i][0][0], points1[i][0][1]))
rec1 = cv2.getRectSubPix(img2, (10, 10), (points2[i][0][0], points2[i][0][1]))
res = cv2.matchTemplate(rec0, rec1, eval('cv2.TM_CCOEFF_NORMED'))
str = np.getbuffer(res)
io = StringIO.StringIO(str)
testnum = chelper.getHelpNumber(io.getvalue())
self.err[i] = testnum
else:
self.err[i] = 0.0
i += 1
def corner_featureVector(corners, image, block_size):
if DEBUG == 1:
print sys._getframe().f_code.co_name
# convert image1 to gray
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
gray = np.float32(gray)
win_w = block_size * 2 + 1
win_h = block_size * 2 + 1
# column means: -180 -135 -90 -45 0 45 90 135 180
# row means: corner
degree_histogram = np.zeros((len(corners), 9, 1), dtype=np.uint8)
for i in range(0, len(corners)):
im = cv2.getRectSubPix(gray, (win_w, win_h), (corners[i][0], corners[i][1]))
# 1st derivative of image
dx = cv2.Sobel(im, cv2.CV_64F, 1, 0, ksize=5)
dy = cv2.Sobel(im, cv2.CV_64F, 0, 1, ksize=5)
angle = np.arctan2(dy, dx)
# range from -180 to 180
# -180 <= angle < -135
# -135 <= angle < -90
# -90 <= angle < -45
# ''''''
for j in range(0, 9):
degree_histogram[i][j] = (((math.pi / 4 * (j - 4)) <= angle) & (angle < (math.pi / 4 * (j - 3)))).sum()
return degree_histogram
def extract_features_from_img(img, kmeans, patch_size=7):
k = kmeans.get_params()['n_clusters']
h,w,c = img.shape
offset = patch_size/2
vectors = []
indices = []
for y in xrange(offset, h-offset):
for x in xrange(offset, w-offset):
patch = cv2.getRectSubPix(img, (patch_size, patch_size), (x,y))
vector = vectorizer.extract_vector_from_patch(patch)
vectors.append(vector)
idx = (2*(2*y/h)+(2*x/w))*k
indices.append(idx)
feature = [0 for _ in xrange(4*k)]
distances = kmeans.transform(vectors)
for dist, idx in zip(distances, indices):
bin = dist.argmin()
min_dist = dist.min()
mean = dist.mean()
strength = max(0, mean - min_dist)
feature[idx+bin] += strength
feature = numpy.array(feature)
feature /= feature.mean()
feature /= feature.std()
return feature
def render(img,sqNum): print 'in render' copy = img.copy() #resize the copy #while (copy.shape[0]>900 or copy.shape[1]>900): # copy = cv2.pyrDown(copy) #paint the grid copyY,copyX = copy.shape[:2] for i in range(1,sqNum-1): ptX=i*copyX/sqNum cv2.line(copy, (ptX,0), (ptX,copyY), (255,0,0), thickness=1) ptY=i*copyY/sqNum cv2.line(copy, (0,ptY), (copyX,ptY), (255,0,0), thickness=1) #get the histograms of the corresponding tiles l = copyX/(2*sqNum) h = copyY/(2*sqNum) squares = [] for i in range(0,sqNum-1): aux = [] for j in range(0,sqNum-1): center = (i*copyX/sqNum+l,j*copyY/sqNum+h) aux.append(getHist(cv2.getRectSubPix(img, (l,h), center))[0]) squares.append(aux) return copy,squares
def fixedPositonBySignet(srcImg):
h, w = srcImg.shape[:2]
img = cv2.cvtColor(srcImg, cv2.COLOR_RGB2HSV)
img = cv2.inRange(img, np.array((90, 60, 50), dtype=np.uint8), np.array((140, 255, 255), dtype=np.uint8))
contours0, hierarchy = cv2.findContours(img, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
for cnt in contours0:
if len(cnt) < 5:
continue
ellipse = cv2.fitEllipse(cnt)
center, r, ang = ellipse
# 角度精确读不足，使用直线检测来确定偏转角度
chileIm = cv2.getRectSubPix(srcImg, (400, 100), center)
new_img = cv2.cvtColor(chileIm, cv2.COLOR_RGB2GRAY)
new_img = cv2.GaussianBlur(new_img, (3, 3), 0)
edges = cv2.Canny(new_img, 50, 150, apertureSize=3)
lines = cv2.HoughLines(edges, 1, np.pi / 180, 200)
if lines is not None:
for line in lines[0]:
rho = line[0] # 第一个元素是距离rho
theta = line[1] # 第二个元素是角度theta
if (theta > (np.pi / 4.)) or (theta < (3.*np.pi / 4.0)):
# 该直线与第一列的交点
pt1 = (0, int(rho / np.sin(theta)))
# 该直线与最后一列的交点
pt2 = (chileIm.shape[1], int((rho - chileIm.shape[1] * np.cos(theta)) / np.sin(theta)))
M = (pt2[1] - pt1[1]) * 1.0 / (pt2[0] - pt1[0])
pi_angle = math.atan(M)
ang = pi_angle * 180 / np.pi
break
else :
ang = ang - 270
if r[0] < 30 or r[1] < 50 or center[1] > h / 2:
continue
rate = r[1] / 175.8019256591797
return center, rate, ang
def __init__(self, frame, rect):
x1, y1, x2, y2 = rect # Extract coordinates of the rectangle to be tracked
w, h = map(cv2.getOptimalDFTSize, [x2 - x1, y2 - y1])
x1, y1 = (x1 + x2 - w) // 2, (y1 + y2 - h) // 2
# 0.5 * x \in [w, h] (-1) = the centre point of the region
self.pos = x, y = x1 + 0.5 * (w - 1), y1 + 0.5 * (h - 1)
self.size = w, h
img = cv2.getRectSubPix(frame, (w, h), (x, y))
# Hanning Window
# http://en.wikipedia.org/wiki/Window_function
# http://en.wikipedia.org/wiki/Window_function#Hann_.28Hanning.29_window
self.win = cv2.createHanningWindow((w, h), cv2.CV_32F)
g = np.zeros((h, w), np.float32)
g[h // 2, w // 2] = 1
g = cv2.GaussianBlur(g, (-1, -1), 2.0)
g /= g.max()
self.G = cv2.dft(g, None, cv2.DFT_COMPLEX_OUTPUT)
self.H1 = np.zeros_like(self.G)
self.H2 = np.zeros_like(self.G)
for i in xrange(128):
a = self.preprocess(rnd_warp(img))
A = cv2.dft(a, None, cv2.DFT_COMPLEX_OUTPUT)
self.H1 += cv2.mulSpectrums(self.G, A, 0, conjB=True)
self.H2 += cv2.mulSpectrums(A, A, 0, conjB=True)
self.update_kernel()
self.update(frame)
self.id = id(self)
def cropImage(edges, blur):
contours, hierarchy = cv2.findContours(edges, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
# borders = find_border_components(contours, edges)
# borders.sort(key=lambda (i, x1, y1, x2, y2): (x2 - x1) * (y2 - y1))
# count=len(borders)
# print len(contours),count
# for i in range(0,count):
# index, left,top,right,bottom =borders[i]
# img_crop=cv2.getRectSubPix(img, (right-left, bottom-top), ((left+right)/2, (top+bottom)/2))
# cv2.imwrite('1right%d' % (left) +'.png', img_crop)
# iand = cv2.bitwise_and(img,img,mask=edges)
# contours, hierarchy = cv2.findContours(edges,cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE)
# cv2.drawContours(edges,contours,-1,(255,255,255),-1)
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (50, 50))
closed = cv2.morphologyEx(edges, cv2.MORPH_CLOSE, kernel)
# closed = cv2.erode(closed, None, iterations = 4)
closed = cv2.dilate(closed, None, iterations=13)
# cv2.imwrite('closed.png',closed)
(cnts, _) = cv2.findContours(closed.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
cnts = sorted(cnts, key=cv2.contourArea, reverse=True)
for i in range(0, 1):
if len(cnts) == 0:
break
if i == 1 and len(cnts) == 1:
break
c = cnts[i]
# compute the rotated bounding box of the largest contour
rect = cv2.minAreaRect(c)
box = np.int0(cv2.cv.BoxPoints(rect))
# cv2.drawContours(img, [box], -1, (0, 255, 0), 3)
left, top, right, bottom = caculateRect(box)
img_crop = cv2.getRectSubPix(blur, (right - left, bottom - top), ((left + right) / 2, (top + bottom) / 2))
cv2.imwrite("right%d" % (left) + ".png", img_crop)
def getCentralRect(img): y = img.shape[0] x = img.shape[1] center = (int(x/2),int(y/2)) patchSize = (int(x/3),int(y/3)) return cv2.getRectSubPix(img, patchSize, center)
def processOneLineImage(gray_img, iTag):
(_, img) = cv2.threshold(gray_img, 110, 255, cv2.THRESH_BINARY_INV)
img = img[:, 2 : img.shape[1] - 2]
scale = psegutils.estimate_scale(img)
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 1))
closed = cv2.dilate(img, kernel, iterations=1)
edges = cv2.Canny(closed, 60, 300)
contours, hierarchy = cv2.findContours(edges, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
cv2.drawContours(edges, contours, -1, (255, 255, 255), 1)
# cv2.imwrite('edges%s.png' % iTag,edges)
boxmap = psegutils.compute_boxmap(img, scale, threshold=(0.4, 10), dtype="B")
# combineBoxmap(boxmap)
cv2.imwrite("box%s.png" % iTag, boxmap * 255)
h_projection = hprojection(boxmap * 255)
top, bottom = cropProjection(h_projection)
regions = splitProjection(h_projection, top, bottom, 30, 2)
# print iTag, top,bottom
# print regions
# print v_projection[1270:1450]
if len(iTag) == 0:
return regions, top, bottom
for region in regions:
topStart, TopEnd = region
cr_img = cv2.getRectSubPix(
gray_img, (gray_img.shape[1] - 4, TopEnd - topStart + 8), (gray_img.shape[1] / 2, (TopEnd + topStart) / 2)
)
cv2.imwrite("%sx%d.png" % (iTag, topStart), cr_img)
return regions, top, bottom
def __init__(self, frame, rect, number):
self.num = number
x1, y1, x2, y2 = rect
w, h = map(cv2.getOptimalDFTSize, [x2-x1, y2-y1])
x1, y1 = (x1+x2-w)//2, (y1+y2-h)//2
self.pos = x, y = x1+0.5*(w-1), y1+0.5*(h-1)
self.size = w, h
img = cv2.getRectSubPix(frame, (w, h), (x, y))
self.win = cv2.createHanningWindow((w, h), cv2.CV_32F)
g = np.zeros((h, w), np.float32)
g[h//2, w//2] = 1
g = cv2.GaussianBlur(g, (-1, -1), 2.0)
g /= g.max()
self.G = cv2.dft(g, flags=cv2.DFT_COMPLEX_OUTPUT)
self.H1 = np.zeros_like(self.G)
self.H2 = np.zeros_like(self.G)
for i in xrange(128):
a = self.preprocess(MOSSE.rnd_warp(img))
A = cv2.dft(a, flags=cv2.DFT_COMPLEX_OUTPUT)
self.H1 += cv2.mulSpectrums(self.G, A, 0, conjB=True)
self.H2 += cv2.mulSpectrums( A, A, 0, conjB=True)
self.update_kernel()
self.update(frame)
def cutTaxPayer(img):
new_img = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)
new_img = cv2.adaptiveThreshold(new_img, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY_INV, 9 , 9)
contours0, hierarchy = cv2.findContours(new_img, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
H, W = img.shape[:2]
result = np.zeros((H, W), np.uint8)
for i in range(len(contours0) - 1, -1, -1):
cnt = contours0[i]
x, y, w, h = cv2.boundingRect(cnt)
if w < 10 or h < 10:
contours0.pop(i)
else :
cv2.rectangle(result, (x, y), (x + w, y + h), (255), 1)
for h in range(H):
for w in range(W):
if result[h][w] == 255:
cv2.line(result, (0, h), (W, h), (255))
break
contours1, hierarchy1 = cv2.findContours(result, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
# for cnt in contours1:
# x, y, w, h = cv2.boundingRect(cnt)
# cv2.rectangle(img, (x, y), (x + w, y + h), (0, 0, 0), 1)
# print x, y
# cv2.imshow("xxx1", img)
if len(contours1) < 2:
return np.ones((H / 4, W * 3 / 4, 3), np.uint8) * 255
x, y, w, h = cv2.boundingRect(contours1[-2])
conter = (x + w / 2, y + h / 2)
img = cv2.getRectSubPix(img, (w, h), conter)
return img
def imagedistort(img):
'''
distort an image
shift horizontally and vertically
rotated clockwise and anticlockwise
'''
shift_range = 0.05
rotate_range = 0.02
h, w = img.shape
size = max(w, h) *2
normal = 255 * np.ones((size, size), np.uint8)
normal[(size - h) / 2: (size + h) / 2, (size - w) / 2: (size + w) / 2] = img
# rotate
degree = 90 * random.uniform(-rotate_range, rotate_range)
M = cv2.getRotationMatrix2D((size/2, size/2), degree, 1)
rotated = cv2.warpAffine(normal, M, (size, size))
# shift
shift_value_x = size / 2 * random.uniform(-shift_range, shift_range)
shift_value_y = size / 2 * random.uniform(-shift_range, shift_range)
M = np.float32([[1, 0, shift_value_x], [0, 1, shift_value_y]])
shift = cv2.warpAffine(rotated, M, (size, size))
# crop
center = (size / 2, size / 2)
crop = cv2.getRectSubPix(shift, (w, h), center)
return crop
def get_sub_image(image, x, y):
dx = float(configs.SIZE-configs.LSTEP-configs.RSTEP) / 15
dy = float(configs.SIZE-configs.BSTEP-configs.TSTEP) / 15
xp = float(configs.LSTEP) + dx * x
yp = float(configs.TSTEP) + dy * y
return cv2.getRectSubPix(image, (int(dx + configs.PATCH_EXPAND), int(dy + configs.PATCH_EXPAND)), (int(xp + dx/2), int(yp + dy/2)))
def __init__(self, frame, rect):
x1, y1, x2, y2 = rect
self.using_lk = False
self.frameWidth = frame.shape[1]
self.frameHeight = frame.shape[0]
w, h = map(cv2.getOptimalDFTSize, [x2-x1, y2-y1])
x1, y1 = (x1+x2-w)//2, (y1+y2-h)//2
self.pos = x, y = x1+0.5*(w-1), y1+0.5*(h-1)
self.size = w, h
self.org_size = w,h
img = cv2.getRectSubPix(frame, (w, h), (x, y))
#self.win = cv2.createHanningWindow((w, h), cv2.CV_32F)
g = np.zeros((h, w), np.float32)
g[h//2, w//2] = 1
g = cv2.GaussianBlur(g, (-1, -1), 2.0)
g /= g.max()
self.G = cv2.dft(g, flags=cv2.DFT_COMPLEX_OUTPUT)
#print "init G",self.G.shape
self.H1 = np.zeros_like(self.G)
self.H2 = np.zeros_like(self.G)
for i in xrange(128):
a = self.preprocess(rnd_warp(img),self.size)
A = cv2.dft(a, flags=cv2.DFT_COMPLEX_OUTPUT)
self.H1 += cv2.mulSpectrums(self.G, A, 0, conjB=True)
self.H2 += cv2.mulSpectrums( A, A, 0, conjB=True)
#print "init imgF:",A.shape
#print "init H1",self.H1.shape
#print "init H2",self.H2.shape
self.update_kernel()
def get_card_texture(card, square=20):
binary = get_binary(card, thresh=150)
contours = find_contours(binary)
if len(contours) < 2:
return None
contour = contours[1]
# get bounding rectangle
rect = cv2.boundingRect(contour)
x, y, w, h = rect
rect = cv2.getRectSubPix(card, (square, square), (x + w / 2, y + h / 2))
gray_rect = cv2.cvtColor(rect, cv2.COLOR_RGB2GRAY)
pixel_std = np.std(gray_rect)
if pixel_std > 4.5:
return sc.PROP_TEXTURE_STRIPED
elif np.mean(gray_rect) > 150:
return sc.PROP_TEXTURE_EMPTY
else:
return sc.PROP_TEXTURE_SOLID
def get_subimage(image, first_anchor, second_anchor):
(fax, fay) = first_anchor
(sax, say) = second_anchor
width = abs(fax - sax) + 1
height = abs(fay - say) + 1
center = ((fax + sax) / 2.0, (fay + say) / 2.0)
subimage = cv2.getRectSubPix(image, (width, height), center)
return subimage
def get_cube_upright():
# Uses the depth image to only take the part of the image corresponding to the closest point and a bit further
global depth_img_avg
global img_bgr8_clean
closest_pnt = np.amin(depth_img_avg)
# resize the depth image so it matches the color one
depth_img_avg = cv2.resize(depth_img_avg, (1280, 960))
# generate a mask with the closest points
img_detection = np.where(depth_img_avg < closest_pnt + val_depth_capture, depth_img_avg, 0)
# put all the pixels greater than 0 to 255
ret, mask = cv2.threshold(img_detection, 0.0, 255, cv2.THRESH_BINARY)
# convert to 8-bit
mask = np.array(mask, dtype=np.uint8)
im2, contours, hierarchy = cv2.findContours(mask, 1, 2, offset=(0, -6))
useful_cnts = list()
uprightrects = list()
img_bgr8_clean_copy = img_bgr8_clean.copy()
for cnt in contours:
if 9000 < cv2.contourArea(cnt) < 15000:
if 420 < cv2.arcLength(cnt, 1) < 560:
useful_cnts.append(cnt)
else:
print("Wrong Lenght 450 < " + str(cv2.arcLength(cnt, 1)) + str(" < 570"))
else:
print ("Wrong Area: 9000 < " + str(cv2.contourArea(cnt)) + " < 15000")
for index, cnts in enumerate(useful_cnts):
min_area_rect = cv2.minAreaRect(cnts) # minimum area rectangle that encloses the contour cnt
(center, size, angle) = cv2.minAreaRect(cnts)
width, height = size[0], size[1]
if not (0.7*height < width < 1.3*height):
print("Wrong Height/Width: " + str(0.7*height) + " < " + str(width) + " < " + str(1.3*height))
continue
points = cv2.boxPoints(min_area_rect) # Find four vertices of rectangle from above rect
points = np.int32(np.around(points)) # Round the values and make it integers
cv2.drawContours(img_bgr8_clean_copy, [points], 0, (0, 0, 255), 2)
cv2.drawContours(img_bgr8_clean_copy, cnts, -1, (255, 0, 255), 2)
cv2.waitKey(1)
# if we rotate more than 90 degrees, the width becomes height and vice-versa
if angle < -45.0:
angle += 90.0
width, height = size[0], size[1]
size = (height, width)
rot_matrix = cv2.getRotationMatrix2D(center, angle, 1.0)
# rotate the entire image around the center of the parking cell by the
# angle of the rotated rect
imgwidth, imgheight = (img_bgr8_clean.shape[0], img_bgr8_clean.shape[1])
rotated = cv2.warpAffine(img_bgr8_clean, rot_matrix, (imgheight, imgwidth), flags=cv2.INTER_CUBIC)
# extract the rect after rotation has been done
sizeint = (np.int32(size[0]), np.int32(size[1]))
uprightrect = cv2.getRectSubPix(rotated, sizeint, center)
uprightrects.append(uprightrect)
uprightrect_copy = uprightrect.copy()
cv2.drawContours(uprightrect_copy, [points], 0, (0, 0, 255), 2)
cv2.imshow('uprightRect ' + str(index), uprightrect_copy)
cv2.imshow('RBG', img_bgr8_clean_copy)
cv2.waitKey(1)
objects_detector(uprightrects)
def extract_vectors_from_img(img,patch_size=7):
img_size = img.shape
offset = patch_size/2
features = []
for y in xrange(offset, img_size[0]-offset):
for x in xrange(offset, img_size[1]-offset):
patch = cv2.getRectSubPix(img, (patch_size, patch_size), (x,y))
features.append(extract_vector_from_patch(patch))
return features
def track(self, frame, pos=None):
(x, y), (w, h) = self.pos, self.size
if pos is not None:
(x,y) = pos
self.last_img = img = cv2.getRectSubPix(frame, (w, h), (x, y))
img = self.preprocess(img)
self.last_resp, (dx, dy), self.psr = self.correlate(img)
self.good = self.psr > 8.0
return np.array([dx, dy])
def subimage(image, centre, theta, width, height):
#output_image = np.zeros((height,width,3), np.uint8)
#mapping = np.array([[np.cos(theta), -np.sin(theta), centre[0]],
# [np.sin(theta), np.cos(theta), centre[1]]])
mapping = cv2.getRotationMatrix2D(centre,theta,1.0)
#map_matrix_cv = cv2.fromarray(mapping)
#cv.GetQuadrangleSubPix(image, output_image, map_matrix_cv)
image = cv2.warpAffine(image,mapping,image[:,:,0].shape,flags=cv2.INTER_LINEAR)
output_image = cv2.getRectSubPix(image,(height,width),centre)
return output_image
def deskew(image, angle):
print angle
image = cv2.bitwise_not(image)
non_zero_pixels = cv2.findNonZero(image)
center, wh, theta = cv2.minAreaRect(non_zero_pixels)
root_mat = cv2.getRotationMatrix2D(center, angle, 1)
rows,cols = image.shape[:2]
rotated = cv2.warpAffine(image, root_mat, (cols, rows), flags=cv2.INTER_CUBIC)
return cv2.bitwise_not(cv2.getRectSubPix(rotated, (cols, rows), center))
def take_photo(argv):
face_found = False
#Which type? Hand or Face
which_cam = 1
name = "face"
if argv:
if argv[0] == "hand":
which_cam = 0
name = "hand"
cascade = cv2.CascadeClassifier(cascade_fn)
cam = create_capture(which_cam)
while True:
ret, img = cam.read()
if name == "face" :
# Do a little preprocessing:
img_copy = cv2.resize(img, (img.shape[1]/2, img.shape[0]/2))
gray = cv2.cvtColor(img_copy, cv2.COLOR_BGR2GRAY)
gray = cv2.equalizeHist(gray)
# Detect the faces (probably research for the options!):
rects = cascade.detectMultiScale(gray)
# Make a copy as we don't want to draw on the original image:
for x, y, width, height in rects:
if 475 < x < 775 :
#cv2.rectangle(img_copy, (int(x-width*0.35), int(y-height/2)), (int(x+width*1.5), int(y+height*1.5)), (255,0,0), 2)
#Save face
cropped = cv2.getRectSubPix(img, (int(width*2)*2, int(height*2)*2), (int(x+width/2)*2, int(y+height/2)*2))
cv2.imwrite("static/captured/" + name + argv[1] + ".png", cropped)
face_found = True
#cv2.imshow('facedetect', cropped)
if cv2.waitKey(20) == 27:
break
if face_found:
break
else:
#920 x 730 at 420, 60
cropped = cv2.getRectSubPix(img, (920,730), (int(420+920/2), int(60+730/2)))
cv2.imwrite("static/captured/" + name + argv[1] + ".png", cropped)
break
def get_neighbors(noisy, sz):
"""
Get sz by sz neighbors and for each pixel value
Note: If grayscale, this will be sz x sz, if RGB, will be sz x sz x 3
Use grayscale
"""
window = (sz, sz)
neighbors = [cv2.getRectSubPix(noisy, window, (y, x)).ravel() \
for x, y in itertools.product(range(noisy.shape[0]), range(noisy.shape[1]))]
neighbors = np.asarray(neighbors)
return (neighbors / 255.0).astype('float32')
def onmouse(event, x, y, flags, param): showImg[:,:,:] = img center = (x,y) pt1 = (x-50,y-50) pt2 = (x+50,y+50) cv2.rectangle(showImg, pt1, pt2, (0,0,255), thickness=2) patch = cv2.pyrUp(cv2.getRectSubPix(img, (100,100), center)) canvas[0:200,0:200,:]=patch canvas[210:466,0:256,:]=getHistogram(patch) if flags & cv2.EVENT_FLAG_LBUTTON: history.append(canvas.copy())
def onmouse(event, x, y, flags, param):
h, w = img.shape[:2]
h1, w1 = small.shape[:2]
x, y = 1.0 * x * h / h1, 1.0 * y * h / h1
zoom = cv2.getRectSubPix(img, (800, 600), (x + 0.5, y + 0.5))
cv2.imshow('zoom', zoom)
index.append(j)
for mask_no in list(set(range(len(validated_masklist))) - set(index)):
final_maskist.append(validated_masklist[mask_no])
for mask in final_maskist:
contour = np.argwhere(mask.transpose() == 255)
rect = cv2.minAreaRect(contour)
width = int(rect[1][0])
height = int(rect[1][1])
centre = (int(rect[0][0]), int(rect[0][1]))
box = cv2.cv.BoxPoints(rect)
box = np.int0(box)
if ((width / float(height) > 1)):
cropped_image = cv2.getRectSubPix(image_mask, (width, height), centre)
else:
cropped_image = cv2.getRectSubPix(image_mask, (height, width), centre)
cropped_image = cv2.cvtColor(cropped_image, cv2.COLOR_BGR2GRAY)
cropped_image = cv2.equalizeHist(cropped_image)
cropped_image = cv2.resize(cropped_image, (260, 63))
cropped_images.append(cropped_image)
#self trained Haar classifier with approx 130 positive images and 50 negative images
number_plate = cv2.CascadeClassifier('output.xml')
index = 0
max_area = 2000
for i in range(len(cropped_images)):
#print('O')
#print(largest[2])
for i in circles[0, :]:
if i[2] > largest[2]:
largest = i
#cv2.circle(whiteMask, (largest[0], largest[1]), largest[2]-5, (0,0,0), -1)
#for i in circles[0,:]:
# cv2.circle(whiteMask, (i[0], i[1]), i[2], (0,0,0), -1)
#whiteMask = cv2.cvtColor(whiteMask,cv2.COLOR_BGR2GRAY)
#masked = cv2.scaleAdd(whiteMask, 1, medianGray)
last = cv2.getRectSubPix(medianGray, (3 * largest[2], 3 * largest[2]),
(largest[0], largest[1]))
_, thr = cv2.threshold(last, 160, 255, cv2.THRESH_BINARY)
Xcoord, Ycoord, Radius = largest[0], largest[1], largest[2] + 10
H, W = thr.shape
# x and y coordinates per every pixel of the image
x, y = np.meshgrid(np.arange(W), np.arange(H))
# squared distance from the center of the circle
d2 = (x - Xcoord)**2 + (y - Ycoord)**2
# mask is True inside of the circle
mask = d2 < Radius**2
draw_circle(frame, largest[2], largest[0], largest[1])
cv2.imshow('frame', frame)
outside = np.ma.masked_where(mask, thr)
width=int(rect[1][0])
height=int(rect[1][1])
centre=(int(rect[0][0]), int(rect[0][1]))
size = (height, width)
angle = int(rect[2])
if angle < -45 :
angle += 90
size = (width,height)
# cv2.drawContours(imgtest,[rect],0,(0,0,255),2)
rot = cv2.getRotationMatrix2D(centre,angle,1)
dim = imgtest.shape
height1 = imgtest.shape[0]
width1 = imgtest.shape[1]
# print dim
orig_rect = cv2.warpAffine(imgtest,rot,(height1,width1))
cropped_image = cv2.getRectSubPix(orig_rect,(height,width),centre)
cropped_image=cv2.cvtColor(cropped_image, cv2.COLOR_BGR2GRAY)
cropped_image=cv2.equalizeHist(cropped_image)
cropped_image = cv2.resize(cropped_image,(260,63))
orig_rects.append(cropped_image)
# showfig(cropped_image,plt.get_cmap('gray'))
images.append(imgtest)
# orig_rect = cv2.resize(orig_rect,(63,260))
# plt.imshow(orig_rect,plt.get_cmap('gray'))
# print size,centre,anglea
# ppc = (8,8)
# cpb = (3,3)
final_images = []
hog_image = []
if (len(orig_rects)>0):
def update(self,current_frame,vis=False):
self.frame_index+=1
old_pos=(np.inf,np.inf)
iter=1
while iter<=self.refinement_iterations and np.any(np.array(old_pos)!=np.array(self._center)):
patch = cv2.getRectSubPix(current_frame,(int(self.base_target_sz[0]*self.sc*(1+self.padding)),
int(self.base_target_sz[1]*self.sc*(1+self.padding))), self._center)
patch=cv2.resize(patch,self.win_sz).astype(np.uint8)
xo_hog,xo_cn= self.get_features(patch,self.cell_size)
xo_cn2, xo_hog2 = self.feature_projection(xo_cn, xo_hog, self.projection_matrix_cn, self.projection_matrix_hog,
self._window)
detect_k_cn=self.dense_gauss_kernel(self.z_cn2,xo_cn2,self.cn_sigma)
detect_k_hog=self.dense_gauss_kernel(self.z_hog2,xo_hog2,self.hog_sigma)
kf=fft2(self.d[0]*detect_k_cn+self.d[1]*detect_k_hog)
responsef=self.alphaf*np.conj(kf)
if self.interpolate_response>0:
if self.interpolate_response==2:
self.interp_sz=(int(self.yf.shape[1]*self.cell_size*self.sc),
int(self.yf.shape[0]*self.cell_size*self.sc))
else:
responsef=self.resize_dft2(responsef,self.interp_sz)
response=np.real(ifft2(responsef))
if vis is True:
self.score = response
self.score = np.roll(self.score, int(np.floor(self.score.shape[0] / 2)), axis=0)
self.score = np.roll(self.score, int(np.floor(self.score.shape[1] / 2)), axis=1)
self.crop_size=self.win_sz
row,col=np.unravel_index(np.argmax(response, axis=None),response.shape)
disp_row=np.mod(row+np.floor((self.interp_sz[1]-1)/2),self.interp_sz[1])-np.floor((self.interp_sz[1]-1)/2)
disp_col=np.mod(col+np.floor((self.interp_sz[0]-1)/2),self.interp_sz[0])-np.floor((self.interp_sz[0]-1)/2)
if self.interpolate_response==0:
translation_vec=list(np.array([disp_row,disp_col])*self.cell_size*self.sc)
elif self.interpolate_response==1:
translation_vec=list(np.array([disp_row,disp_col])*self.sc)
elif self.interpolate_response==2:
translation_vec=[disp_row,disp_col]
trans=np.sqrt(self.win_sz[0]*self.win_sz[1])*self.sc/3
old_pos=self._center
self._center=(old_pos[0]+translation_vec[1],old_pos[1]+translation_vec[0])
iter+=1
self.sc = self.scale_estimator.update(current_frame, self._center, self.base_target_sz,
self.sc)
if self.scale_type == 'normal':
self.sc = np.clip(self.sc, a_min=self._min_scale_factor,
a_max=self._max_scale_factor)
patch = cv2.getRectSubPix(current_frame, (int(self.base_target_sz[0] * self.sc * (1 + self.padding)),
int(self.base_target_sz[1] * self.sc * (1 + self.padding))),
self._center)
patch = cv2.resize(patch, self.win_sz).astype(np.uint8)
xo_hog,xo_cn=self.get_features(patch,self.cell_size)
self.z_hog=(1-self.lr_hog)*self.z_hog+self.lr_hog*xo_hog
self.z_cn=(1-self.lr_cn)*self.z_cn+self.lr_cn*xo_cn
data_matrix_cn = self.z_cn.reshape((-1, self.z_cn.shape[2]))
pca_basis_cn, _, _ = np.linalg.svd(data_matrix_cn.T.dot(data_matrix_cn))
self.projection_matrix_cn = pca_basis_cn[:, :self.num_compressed_dim_cn]
data_matrix_hog = self.z_hog.reshape((-1, self.z_hog.shape[2]))
pca_basis_hog, _, _ = np.linalg.svd(data_matrix_hog.T.dot(data_matrix_hog))
self.projection_matrix_hog = pca_basis_hog[:, :self.num_compressed_dim_hog]
self.z_cn2, self.z_hog2 = self.feature_projection(self.z_cn, self.z_hog, self.projection_matrix_cn, self.projection_matrix_hog,
self._window)
if self.frame_index%self.modnum==0:
self.train_model()
target_sz=((self.base_target_sz[0]*self.sc),(self.base_target_sz[1]*self.sc))
return [(self._center[0] - target_sz[0] / 2), (self._center[1] - target_sz[1] / 2), target_sz[0],target_sz[1]]
def ExtractTile(self):
height, width, depth = self.tile.shape
# Contour on Frame Boundary
img = np.zeros((height, width, 3), np.uint8)
cv2.rectangle(img, (0, 0), (width, height), (0, 0, 255), 2)
contour_image = cv2.cvtColor(img, cv2.cv.CV_BGR2GRAY)
contours_frame, hierarchy = cv2.findContours(
contour_image, cv2.cv.CV_RETR_EXTERNAL,
cv2.cv.CV_CHAIN_APPROX_SIMPLE)
#cv2.drawContours(img, contours_frame, -1, (0, 255, 0), 3 )
#cv2.imshow("Contour_Frame",img)
#Contour on Tiles
grey_image = cv2.cvtColor(self.tile, cv2.cv.CV_BGR2GRAY)
thresh = cv2.threshold(grey_image, 0, 255,
cv2.cv.CV_THRESH_BINARY + cv2.cv.CV_THRESH_OTSU)
canny_image = cv2.Canny(grey_image, thresh[0], thresh[0] * 3)
contours, hierarchy = cv2.findContours(canny_image,
cv2.cv.CV_RETR_EXTERNAL,
cv2.cv.CV_CHAIN_APPROX_NONE)
length = len(contours) # check for tile-less frames
if length:
# Find the index of the largest contour
areas = [cv2.contourArea(c) for c in contours]
#areas = [cv2.arcLength(c,True) for c in contours]
max_index = np.argmax(areas)
#cv2.drawContours(self.tile,contours[max_index], -1, (255, 0, 0), 3 )
#cv2.imshow("Contour_Approx", self.tile)
# To find indices of rotated rectangle
bound_points = cv2.minAreaRect(contours[max_index])
#print (type(bound_points)) -->((c0,c1),(w,h),theta)
#print bound_points[0][0],bound_points[0][1],bound_points[1][0],bound_points[1][1]
#print bound_points[2]
c0 = bound_points[0][0]
c1 = bound_points[0][1]
w = bound_points[1][0]
h = bound_points[1][1]
theta = (bound_points[2])
# Tile Boundness - In Frame - Test
self.boundnessflag = 1
self.areaflag = 1
# Test For maxtile area
if (self.firstframe == True):
self.MaxAreaTile = areas[max_index]
self.firstframe = 0
else:
if (areas[max_index] < self.MaxAreaTile):
self.areaflag = 0
else:
self.areaflag = 1
self.MaxAreaTile = areas[max_index]
# Test Using boundingRect
x_br, y_br, w_br, h_br = cv2.boundingRect(contours[max_index])
#print x_br,y_br,w_br,h_br
ret_val_1 = cv2.pointPolygonTest(contours_frame[0], (x_br, y_br),
False)
ret_val_2 = cv2.pointPolygonTest(contours_frame[0],
(x_br, (y_br + w_br)), False)
ret_val_3 = cv2.pointPolygonTest(contours_frame[0],
((x_br + h_br), y_br), False)
ret_val_4 = cv2.pointPolygonTest(contours_frame[0],
((x_br + h_br), (y_br + w_br)),
False)
#print ret_val_1,ret_val_2,ret_val_3,ret_val_4
if ((ret_val_1 == 0) or (ret_val_2 == 0) or (ret_val_3 == 0)
or (ret_val_4 == 0)):
self.boundnessflag = 0
#cv2.rectangle(tile,(x_br,y_br),(x_br+w_br,y_br+h_br),(255,0,0),2)
#cv2.imshow("Boundrect", tile)
# Test Using approxPolyDP
"""
approx = cv2.approxPolyDP(contours[max_index],0.1*cv2.arcLength(contours[max_index],True),True)
#cv2.drawContours(tile, approx, -1, (255, 0, 0), 3 )
cv2.imshow("Contour_Approx", tile)
out = range(len(approx))
print len(approx),type(approx),approx
for i in range(len(approx)):
a,b = approx[i][0]
print a , b
out[i] = cv2.pointPolygonTest(contours_frame[0],(a,b) , False)
print out[i]
if (out[i] == False):
flag = 0
break
"""
# Create output image
#if(c0 and c1 and w and h):
if (self.boundnessflag and self.areaflag):
rot_mat = cv2.getRotationMatrix2D(
(bound_points[0][0], bound_points[0][1]), theta, 1)
#print(type(rot_mat),rot_mat)
rotated = cv2.warpAffine(self.tile, rot_mat, (width, height))
#print rotated
self.final = cv2.getRectSubPix(rotated, ((int)(w), (int)(h)),
(c0, c1))
self.outflag = 1
#cv2.imshow("Final", self.final)
else:
self.outflag = 0
def load_train_valid_images(image_files,
train_index,
batch_index,
batch_size,
rotationAngle,
cropImageSize,
finalImageSize,
channel,
substract_each_image_mean=False,
augmentation_method="on_the_fly"):
"""Since the whole training image set contains 61578 images, and is too large to fit in
the memory, we use this function to load a mini-batch images at a time, and use that
mini-batch to train the CNN. Note that this
:type image_files: list
:param image_files: contains all the file names of the whole training images
NOTE: image_files contains both training set and validation set
:type train_index: array
:param train_index: contains all the indices for the training set
:type batch_index: int
:param batch_index: the index of this batch
:type batch_size: int
:param batch_size: size of each mini-batch
"""
# the index of this mini-batch
train_batch_index = train_index[batch_index *
batch_size:(batch_index + 1) * batch_size]
train_image_files = [image_files[i] for i in train_batch_index]
images_train = np.zeros((batch_size, channel * (finalImageSize**2)),
dtype=theano.config.floatX)
for i, imgf in enumerate(train_image_files):
# read JPEG image, see:
# http://docs.opencv.org/modules/highgui/doc/
# reading_and_writing_images_and_video.html?highlight=imread#imread
#
# flags>0 for 3-channel color image
# flags=0 for grayscale image
flags = (0 if channel == 1 else 3)
if augmentation_method == "pre_computed":
s = imgf.split("/")
f, n = s[0], s[1]
imgf = f + "/angle{}".format(np.int32(rotationAngle)) + "/" + n
imgf = imgf[:-4] + "_rotationAngle{}_cropImageSize{}.jpg".format(
rotationAngle, cropImageSize)
finalImage = cv2.imread(imgf, flags)
elif augmentation_method == "on_the_fly":
originalImage = cv2.imread(imgf, flags)
# use only the central part of size patchSize x patchSize
height = originalImage.shape[0]
width = originalImage.shape[1]
center = (height / 2.0, width / 2.0)
# about geometric transform using opencv-python, see
# http://opencv-python-tutroals.readthedocs.org/en/latest/py_tutorials/
# py_imgproc/py_geometric_transformations/py_geometric_transformations.html
#image_counter = 0
#for angle in rotationAngles:
# we do not scale it, so we set scale = 1
rotationMatrix = cv2.getRotationMatrix2D(center,
rotationAngle,
scale=1)
# keep the output image with the same size as the input
rotatedImage = cv2.warpAffine(originalImage, rotationMatrix,
(height, width))
# see,
# http://docs.opencv.org/modules/imgproc/doc/geometric_transformations.html#warpaffine
#print center_coord
# crop the central part of size patchSize
#for cropSize in cropImageSizes:
croppedImage = cv2.getRectSubPix(rotatedImage,
(cropImageSize, cropImageSize),
center)
# resize the central part to size imageSize
finalImage = cv2.resize(croppedImage,
(finalImageSize, finalImageSize))
# whether we substract mean of each image (and each channel for channel = 3)
if (substract_each_image_mean == True):
meanImage = np.mean(np.mean(finalImage, axis=0), axis=0)
finalImage -= meanImage
# swap axes, so dim = (imageSize, imageSize, 3) becomes dim = (3, imageSize, imageSize)
if (channel == 3):
finalImage = np.swapaxes(np.swapaxes(finalImage, 1, 2), 0, 1)
# reshape it into 1-D rasterized image
finalImage = np.reshape(finalImage, channel * (finalImageSize**2))
images_train[i] = finalImage
return (images_train)
heightNew = int(width * fabs(sin(radians(degree))) +
height * fabs(cos(radians(degree))))
widthNew = int(height * fabs(sin(radians(degree))) +
width * fabs(cos(radians(degree))))
matRotation = cv2.getRotationMatrix2D((width / 2, height / 2), degree,
1)
matRotation[0, 2] += (widthNew - width) / 2 # 重点在这步，目前不懂为什么加这步
matRotation[1, 2] += (heightNew - height) / 2 # 重点在这步
img = cv2.warpAffine(img,
matRotation, (widthNew, heightNew),
borderValue=(255, 255, 255))
# crop image(Size patchSize：获取矩形的大小, Point2f center：获取的矩形在原图像中的位置)
# target_img = cv2.getRectSubPix(imgRotation, (900, 1000), (550, 1100)) # go straight
# img = cv2.getRectSubPix(img, (1080, 1200), (500, 1000)) # go straight(wanghaidong)
img = cv2.getRectSubPix(
img, (sub_imgage_width, sub_image_height),
(sub_image_center_x, sub_image_center_y)) # go straight(wuyiqiang)
if not os.path.exists(crop_dest_image_path):
os.makedirs(crop_dest_image_path)
# /home/dong/PycharmProjects/traffic-gesture-recognition/datasets/my/go_straight_full_image
# /home/dong/PycharmProjects/traffic-gesture-recognition/datasets/my/park_right/yangluxing
cv2.imwrite(os.path.join(crop_dest_image_path, file), img)
# cv2.imshow('img', target_img)
# cv2.waitKey()
def get_vein_img(self, save_vein_pic=True, save_bb=True):
crop = []
for sample in range(0, self.total_input):
# Error removing for augmented data---------------------
file, point, point_pred = str(
self.img_name[sample]
), self.output[sample], self.target[sample]
if ((file.find('_flrot_') != -1) |
(file.find('_flrotVera_') != -1)):
point1 = np.array(point[0:2])
point2 = np.array(point[2:4])
point_changed = []
point_changed.append(point2)
point_changed.append(point1)
self.output[sample] = np.array(point_changed).reshape((1, 4))
point1 = np.array(point_pred[0:2])
point2 = np.array(point_pred[2:4])
point_changed = []
point_changed.append(point2)
point_changed.append(point1)
self.target[sample] = np.array(point_changed).reshape((1, 4))
# -------------------------------------------------------
top_left = self.output[sample, 0:2]
top_right = self.output[sample, 2:4]
# Find the angle to rotate the image
angle = (180 / np.pi) * (np.arctan(
(top_left[1] - top_right[1]) / (top_left[0] - top_right[0])))
# Rotate the image to cut rectangle from the images
points_pred = (self.output[sample]).reshape((1, 2, 2))
points_test = (self.target[sample]).reshape((1, 2, 2))
img = cv2.imread(self.data_folder + self.img_name[sample])
# image = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
image = []
image.append(img)
image = np.array(image)
image_rotated, keypoints_pred_rotated = iaa.Affine(rotate=-angle)(
images=image, keypoints=points_pred)
_, keypoints_test_rotated = iaa.Affine(rotate=-angle)(
images=image, keypoints=points_test)
# Check if the image is fully rotated that left goes to the right side of hand
if (keypoints_pred_rotated[0, 0, 0] > keypoints_pred_rotated[0, 1,
0]):
# Again rotate the picture to 180 with the points
image = image_rotated
image_rotated, keypoints_pred_rotated = iaa.Affine(rotate=180)(
images=image, keypoints=keypoints_pred_rotated)
_, keypoints_test_rotated = iaa.Affine(rotate=180)(
images=image, keypoints=keypoints_test_rotated)
image_rotated = image_rotated[0]
keypoints_pred_rotated = keypoints_pred_rotated.reshape((2, 2))
keypoints_test_rotated = keypoints_test_rotated.reshape((2, 2))
# Rotated Points
top_left = keypoints_pred_rotated[0]
top_left[0] = top_left[0] - self.th
top_right = keypoints_pred_rotated[1]
top_right[0] = top_right[0] + self.th
self.width = int(abs(top_right - top_left)[0])
self.height = int(self.width * (90 / 80))
centre = tuple([
top_left[0] + int(self.width / 2),
top_left[1] + int(self.height / 2)
])
# Crop the Vein Image
cropped = cv2.getRectSubPix(image_rotated,
(self.width, self.height), centre)
crop.append(cropped)
if (save_vein_pic):
cv2.imwrite(self.cropped_fldr + self.img_name[sample], cropped)
# Draw Predicted Troughs
points = keypoints_pred_rotated.reshape((2, 2))
color = [(255, 255, 255),
(0, 0, 0)] # Left - White, # Right - Black
count = 0
for point in points:
point = np.array(point).astype(int)
cv2.circle(image_rotated, (point[0], point[1]), 5,
color[count], -1)
count += 1
# Draw Actual Troughs
points = keypoints_test_rotated.reshape((2, 2))
for point in points:
point = np.array(point).astype(int)
cv2.circle(image_rotated, (point[0], point[1]), 5, (255, 0, 0),
-1)
bottom_right = [
int(top_left[0] + self.width),
int(top_left[1] + self.height)
]
# Draw Bounding Boxes and Save the image
image_rotated = cv2.rectangle(image_rotated, tuple(top_left),
tuple(bottom_right), (0, 0, 0), 2)
if (save_bb):
cv2.imwrite(self.bounding_box_folder + self.img_name[sample],
image_rotated)
crop = np.array(crop)
return crop
triangle_hypotenus = np.linalg.norm(
np.array([sorted_chars[0]['cx'], sorted_chars[0]['cy']]) -
np.array([sorted_chars[-1]['cx'], sorted_chars[-1]['cy']]))
angle = np.degrees(np.arcsin(triangle_height / triangle_hypotenus))
rotation_matrix = cv2.getRotationMatrix2D(center=(plate_cx, plate_cy),
angle=angle,
scale=1.0)
img_rotated = cv2.warpAffine(img_thresh,
M=rotation_matrix,
dsize=(width, height))
img_cropped = cv2.getRectSubPix(img_rotated,
patchSize=(int(plate_width),
int(plate_height)),
center=(int(plate_cx), int(plate_cy)))
if img_cropped.shape[1] / img_cropped.shape[
0] < MIN_PLATE_RATIO or img_cropped.shape[1] / img_cropped.shape[
0] < MIN_PLATE_RATIO > MAX_PLATE_RATIO:
continue
plate_imgs.append(img_cropped)
plate_infos.append({
'x': int(plate_cx - plate_width / 2),
'y': int(plate_cy - plate_height / 2),
'w': int(plate_width),
'h': int(plate_height)
})
def add_new_cube(self, image):
""""This function takes care of adding a new colorcube to the range of detection cubes."""
# Try to add the area specified by the most recent click.
# Determine the area to use.
# This is a square area of self.cube_size pixels.
area_to_be_added = cv2.getRectSubPix(
image, (self.cube_size, self.cube_size),
(float(self.add_x), float(self.add_y)))
# Determine the min and max values in that area.
# Split the image
(hue, saturation, value) = cv2.split(area_to_be_added)
# Determine the min and max in the area.
# These min and max are increased/decreased to improve detection.
hue_max = numpy.amin([180, numpy.amax(hue) + self.increase_h])
hue_min = numpy.amax([0, numpy.amin(hue) - self.increase_h])
print hue
if (hue_max - hue_min) > 100:
#Ok, probably it is some kind of a red color (around 0):
#So we need to determin the min and max around 0:
#TODO: Make more efficient using numpy or so:
hue_min = 180
hue_max = 0
for hue_val_line in hue:
for hue_val in hue_val_line:
print hue_val
if hue_val > 90:
if hue_val < hue_min:
hue_min = hue_val
if hue_val <= 90:
if hue_val > hue_max:
hue_max = hue_val
saturation_max = numpy.amin(
[255, numpy.amax(saturation) + self.increase_s])
saturation_min = numpy.amax(
[0, numpy.amin(saturation) - self.increase_s])
value_max = numpy.amin([255, numpy.amax(value) + self.increase_v])
value_min = numpy.amax([0, numpy.amin(value) - self.increase_v])
# Add these to a dict.
new_cube = {
'h_lower': hue_min,
'h_upper': hue_max,
's_lower': saturation_min,
's_upper': saturation_max,
'v_lower': value_min,
'v_upper': value_max
}
print "Made new dict."
print new_cube
# Add the dict to the detection dicts.
((self.to_be_detected[self.current_target])[1]).append(
copy.deepcopy(new_cube))
# And it's done!
self.add_from_next_image = False
INIT_LR = 1e-3
BS = 32
#intialize the data and labels
print("[INFO] loading images...")
data = []
labels = []
# grab the image ad randomly shuffle them
imagePaths = sorted(list(paths.list_images(args["dataset"])))
random.seed(42)
random.shuffle(imagePaths)
# lopp over the input images
for imagePath in imagePaths:
# load the image ,pre-process it , and store it to the data list
image = cv2.imread(imagePath)
image = cv2.getRectSubPix(image, (320, 240), (150, 150))
thresh = 136
image = cv2.threshold(image, thresh, 255, cv2.THRESH_BINARY)[1]
image = cv2.resize(image, (28, 28))
image = img_to_array(image)
data.append(image)
# extract the class label from the image path and update the label list
label = imagePath.split(os.path.sep)[-2]
if label == "no":
label = 0
elif label == "first":
label = 1
elif label == "second":
label = 2
elif label == "third":
def init(self,first_frame,bbox):
bbox = np.array(bbox).astype(np.int64)
x0, y0, w, h = tuple(bbox)
self.target_sz=(w,h)
self._center = (int(x0 + w / 2),int( y0 + h / 2))
if w*h>self.translation_model_max_area:
self.sc=np.sqrt(w*h/self.translation_model_max_area)
else:
self.sc=1.
self.base_target_sz=(w/self.sc,h/self.sc)
self.win_sz = (int(np.floor(self.base_target_sz[0] * (1 + self.padding))), int(np.floor(self.base_target_sz[1] * (1 + self.padding))))
output_sigma=np.sqrt(self.base_target_sz[0]*self.base_target_sz[1])*self.output_sigma_factor/self.cell_size
use_sz=(int(np.floor(self.win_sz[0]/self.cell_size)),int(np.floor(self.win_sz[1]/self.cell_size)))
self.yf = fft2(0.5*gaussian2d_rolled_labels(use_sz,sigma=output_sigma))
self.interp_sz=(use_sz[0]*self.cell_size,use_sz[1]*self.cell_size)
self._window=cos_window(use_sz)
if self.scale_type=='normal':
self.scale_estimator = DSSTScaleEstimator(self.target_sz, config=self.scale_config)
self.scale_estimator.init(first_frame, self._center, self.base_target_sz, self.sc)
self._num_scales = self.scale_estimator.num_scales
self._scale_step = self.scale_estimator.scale_step
self._min_scale_factor = self._scale_step ** np.ceil(
np.log(np.max(5 / np.array(([self.win_sz[0], self.win_sz[1]])))) / np.log(self._scale_step))
self._max_scale_factor = self._scale_step ** np.floor(np.log(np.min(
first_frame.shape[:2] / np.array([self.base_target_sz[1], self.base_target_sz[0]]))) / np.log(
self._scale_step))
elif self.scale_type=='LP':
self.scale_estimator=LPScaleEstimator(self.target_sz,config=self.scale_config)
self.scale_estimator.init(first_frame,self._center,self.base_target_sz,self.sc)
self.cn_sigma = self.cn_sigma_color
self.hog_sigma = self.hog_sigma_color
self.lr_hog = self.lr_hog_color
self.lr_cn = self.lr_cn_color
self.modnum = self.gap
self.is_gray = False
patch=cv2.getRectSubPix(first_frame,self.win_sz,self._center).astype(np.uint8)
self.z_hog,self.z_cn=self.get_features(patch,cell_size=self.cell_size)
data_matrix_cn=self.z_cn.reshape((-1,self.z_cn.shape[2]))
pca_basis_cn,_,_=np.linalg.svd(data_matrix_cn.T.dot(data_matrix_cn))
self.projection_matrix_cn=pca_basis_cn[:,:self.num_compressed_dim_cn]
data_matrix_hog=self.z_hog.reshape((-1,self.z_hog.shape[2]))
pca_basis_hog,_,_=np.linalg.svd(data_matrix_hog.T.dot(data_matrix_hog))
self.projection_matrix_hog=pca_basis_hog[:,:self.num_compressed_dim_hog]
self.z_cn2,self.z_hog2=self.feature_projection(self.z_cn,self.z_hog,self.projection_matrix_cn,self.projection_matrix_hog,
self._window)
self.frame_index=1
self.d=self.train_model()
def showresultrect(img_rotated, dsize, center):
img_crop = cv2.getRectSubPix(img_rotated, (dsize[0], dsize[1]), center)
img_test = cv2.resize(img_crop, (136, 36))
return img_test, img_crop
fDistToWx = interp1d(distance_arr, x_interval)
fDistToWz = interp1d(distance_arr, fWxToWz(x_interval))
aspect = height / width
output_width = 500
output_height = int(aspect * output_width)
print 'Begin Unwarping'
output_image = np.zeros((output_height, output_width, 3))
for oy in range(len(output_image)):
for ox in range(len(output_image[0])):
d = (ox / float(output_width)) * width
v = (oy / float(output_height)) * height
wx = fDistToWx(d)
wy = v
wz = fDistToWz(d)
ix, iy = world_to_source(wx, wy, wz)
image_x = (image_width / 2) - ix
image_y = (image_height / 2) - iy
pixel = cv2.getRectSubPix(image, (1, 1), (image_x, image_y))
output_image[output_height - oy - 1,
output_width - ox - 1] = pixel[0, 0] / 255.0
cv2.imwrite('./test_focal_length/' + str(focal) + '.png', output_image * 255.0)
cv.circle(blobs, (x, y), int(radius), color, 5)
cv.circle(blobs, (x, y), int(1), color, 5)
draw_str(blobs, (x, y), "%d" % i)
cv.imshow('blobs', blobs)
cv.imshow('index_map', index_map)
s = 4
patchsize = (s, s)
# triangle center
# TODO: get from tracker
px = 600 * f
py = 200 * f
center = (px, py)
zoom = cv.getRectSubPix(hsv1, patchsize, center)
show('zoom', zoom)
# t = 64
t = sz // s
# t*t = total patches
i = read_curr_index
ci = ((i % (t * t)) // t) * s
cj = (i % t) * s
# zoom[:,:,1] = 255 # s
# zoom[:,:,2] = 255 # v
rgb_zoom = cv.cvtColor(zoom, cv.COLOR_HSV2BGR)
img[ci:ci + s, cj:cj + s, :] = rgb_zoom
def stitch(img1, img2, H, inv_H):
print('----Stitching Images ------- ')
# Find StitchedImage size
# print('Image1 dimensions', img1.shape)
# Four Points of Image1
h1, w1, d1 = img1.shape
img1points1 = [[0, 0]]
img1points2 = [[w1, 0]]
img1points3 = [[0, h1]]
img1points4 = [[w1, h1]]
# points1 = np.array([img1points1, img1points2, img1points3, img1points4], np.float32)
# print('Points1 Image', points1)
# Four points of Image2
h2, w2, d2 = img2.shape
x, y = project(0, 0, inv_H)
x1, y1 = project(w2, 0, inv_H)
x2, y2 = project(0, h2, inv_H)
x3, y3 = project(w2, h2, inv_H)
img2points1 = [[x, y]]
img2points2 = [[x1, y1]]
img2points3 = [[x2, y2]]
img2points4 = [[x3, y3]]
points = np.array(
[img1points1, img1points2, img1points3, img1points4, img2points1, img2points2, img2points3, img2points4],
np.float32)
boundaryPoints = cv2.boundingRect(points)
w = boundaryPoints[2] - boundaryPoints[0]
h = boundaryPoints[3] - boundaryPoints[1]
stichedImage = np.zeros([h, w, 3], np.uint8)
# cv2.imshow('Stitched Window', stichedImage)
# cv2.waitKey()
print('Image 1 dimensions', img1.shape, 'Image2 dimensions', img2.shape)
print('Image 1 boundary points', img1points1, img1points2, img1points3, img1points4)
print('Image 2 boundary', (0, 0), (w2, 0), (0, h2), (w2, h2))
print('Projected Point', img2points1, img2points2, img2points3, img2points4)
print('Boundary Points', boundaryPoints)
print('Stitched Image', stichedImage.shape)
# Copy Image 1 to stichedImage
# stichedImage[0:img1.shape[0], 0:img1.shape[1]] = img1
for y in range(0, img1.shape[0]):
for x in range(0, img1.shape[1]):
stichedImage[y - boundaryPoints[1], x - boundaryPoints[0]] = img1[y,x]
# Project the stitchedImage in image2 space
'''
1. Project each pixel in Stitched Image to image2
2. IF the point lies within image2 boundaries add pixel to stitchedImage
'''
for y in range(boundaryPoints[1], stichedImage.shape[0]):
for x in range(boundaryPoints[0], stichedImage.shape[1]):
x1, y1 = project(x,y, H[0])
# print('Projected Points', x1, y1)
if(0 <= x1 < img2.shape[1]) and (0 <= y1 < img2.shape[0]):
pixelValueImg2 = cv2.getRectSubPix(img2,(1,1), (x1, y1) )
if (y-boundaryPoints[1] < stichedImage.shape[0]) and (x-boundaryPoints[0]< stichedImage.shape[1]):
stichedImage[y - boundaryPoints[1], x - boundaryPoints[0]] = pixelValueImg2[0][0]
print('-------Exiting the Stitched Image -----------')
return stichedImage
def extractPlate(imgOriginal, listOfMatchingChars):
possiblePlate = PossiblePlate.PossiblePlate(
) # this will be the return value
listOfMatchingChars.sort(
key=lambda matchingChar: matchingChar.intCenterX
) # sort chars from left to right based on x position
# calculate the center point of the plate
fltPlateCenterX = (
listOfMatchingChars[0].intCenterX +
listOfMatchingChars[len(listOfMatchingChars) - 1].intCenterX) / 2.0
fltPlateCenterY = (
listOfMatchingChars[0].intCenterY +
listOfMatchingChars[len(listOfMatchingChars) - 1].intCenterY) / 2.0
ptPlateCenter = fltPlateCenterX, fltPlateCenterY
# calculate plate width and height
intPlateWidth = int(
(listOfMatchingChars[len(listOfMatchingChars) - 1].intBoundingRectX +
listOfMatchingChars[len(listOfMatchingChars) - 1].intBoundingRectWidth
- listOfMatchingChars[0].intBoundingRectX) *
PLATE_WIDTH_PADDING_FACTOR)
intTotalOfCharHeights = 0
for matchingChar in listOfMatchingChars:
intTotalOfCharHeights = intTotalOfCharHeights + matchingChar.intBoundingRectHeight
# end for
fltAverageCharHeight = intTotalOfCharHeights / len(listOfMatchingChars)
intPlateHeight = int(fltAverageCharHeight * PLATE_HEIGHT_PADDING_FACTOR)
# calculate correction angle of plate region
fltOpposite = listOfMatchingChars[
len(listOfMatchingChars) -
1].intCenterY - listOfMatchingChars[0].intCenterY
fltHypotenuse = DetectChars.distanceBetweenChars(
listOfMatchingChars[0],
listOfMatchingChars[len(listOfMatchingChars) - 1])
fltCorrectionAngleInRad = math.asin(fltOpposite / fltHypotenuse)
fltCorrectionAngleInDeg = fltCorrectionAngleInRad * (180.0 / math.pi)
# pack plate region center point, width and height, and correction angle into rotated rect member variable of plate
possiblePlate.rrLocationOfPlateInScene = (tuple(ptPlateCenter),
(intPlateWidth, intPlateHeight),
fltCorrectionAngleInDeg)
# final steps are to perform the actual rotation
# get the rotation matrix for our calculated correction angle
rotationMatrix = cv2.getRotationMatrix2D(tuple(ptPlateCenter),
fltCorrectionAngleInDeg, 1.0)
height, width, numChannels = imgOriginal.shape # unpack original image width and height
imgRotated = cv2.warpAffine(imgOriginal, rotationMatrix,
(width, height)) # rotate the entire image
imgCropped = cv2.getRectSubPix(imgRotated, (intPlateWidth, intPlateHeight),
tuple(ptPlateCenter))
possiblePlate.imgPlate = imgCropped # copy the cropped plate image into the applicable member variable of the possible plate
return possiblePlate
def cropImage(image, rectangle):
x, y = rectangle[0], rectangle[1]
h, w = rectangle[2], rectangle[3]
return cv2.getRectSubPix(image, (int(h), int(w)), (x + h / 2, y + w / 2))
# pack plate region center point, width and height, and correction angle into rotated rect member variable of plate
possiblePlate.rrLocationOfPlateInScene = (tuple(plateCenter),
(plateWidth, plateHeight),
correctionAngleInDeg)
# get the rotation matrix for our calculated correction angle
rotationMatrix = cv2.getRotationMatrix2D(tuple(plateCenter),
correctionAngleInDeg, 1.0)
height, width, numChannels = img.shape
# rotate the entire image
imgRotated = cv2.warpAffine(img, rotationMatrix, (width, height))
# crop the image/plate detected
imgCropped = cv2.getRectSubPix(imgRotated, (plateWidth, plateHeight),
tuple(plateCenter))
# copy the cropped plate image into the applicable member variable of the possible plate
possiblePlate.Plate = imgCropped
# populate plates_list with the detected plate
if possiblePlate.Plate is not None:
plates_list.append(possiblePlate)
# draw a ROI on the original image
for i in range(0, len(plates_list)):
# finds the four vertices of a rotated rect - it is useful to draw the rectangle.
p2fRectPoints = cv2.boxPoints(plates_list[i].rrLocationOfPlateInScene)
# roi rectangle colour
rectColour = (0, 255, 0)
def extract_patch_sz(img, sz, xy):
sub = cv2.getRectSubPix(img, (int(round(sz)), int(round(sz))), xy)
res = cv2.resize(sub, (input_sz, input_sz))
return res
cv2.imwrite('./results/result_phase_1.png', canvas)
relighted = face_relighting.relight(image, is_bgr=True)
cv2.imwrite('./results/result_phase_2.png', relighted)
mask = removeBG_API_request.remove_background(relighted,
image_format='.jpg',
is_RGB=False)[:, :, 3]
mask = cv2.cvtColor(mask, cv2.COLOR_GRAY2BGR)
cv2.imwrite('./results/result_phase_3.png', mask)
h_div_w = 1.231
new_w = w
new_h = int(w * h_div_w)
output = cv2.getRectSubPix(relighted, (new_w, new_h), (w / 2, h / 2))
mask = cv2.getRectSubPix(mask, (new_w, new_h), (w / 2, h / 2))
white = np.ones_like(output) * 255
result = (white * (1 - (mask / 255.)) + (mask / 255.) * output).astype('uint8')
cv2.imwrite('./results/result_last_phase.png', result)
fig, axs = plt.subplots(3, 3)
fig.suptitle('Diego Bonilla auto-photo-ID')
axs[0, 0].imshow(white)
axs[0, 0].axis('off')
axs[2, 0].imshow(white)
axs[2, 0].axis('off')
def get_subpixel(img, y, x):
patch = cv2.getRectSubPix(img, (1, 1), (x, y))
return patch[0][0]
def get_license_plate_char(self, image):
# Read image
img_ori = image
if type(image) == str:
if platform.system().lower() == 'windows' and image[0] == '~':
image = os.environ['USERPROFILE'] + image[1:]
image = os.path.abspath(image)
img_ori = cv2.imread(image)
if type(img_ori) is not np.ndarray:
raise ValueError('ERROR: invalid image!')
height, width, channel = img_ori.shape
# Convert image to grayscale
gray = cv2.cvtColor(img_ori, cv2.COLOR_BGR2GRAY)
# Maximize constrast
structuringElement = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
imgTopHat = cv2.morphologyEx(gray, cv2.MORPH_TOPHAT,
structuringElement)
imgBlackHat = cv2.morphologyEx(gray, cv2.MORPH_BLACKHAT,
structuringElement)
imgGrayscalePlusTopHat = cv2.add(gray, imgTopHat)
gray = cv2.subtract(imgGrayscalePlusTopHat, imgBlackHat)
# Thresholding
img_blurred = cv2.GaussianBlur(gray, ksize=(5, 5), sigmaX=0)
usingAdaptive = True
if usingAdaptive:
img_thresh = cv2.adaptiveThreshold(
img_blurred,
maxValue=255.0,
adaptiveMethod=cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
thresholdType=cv2.THRESH_BINARY_INV,
blockSize=19,
C=12)
else:
_, img_thresh = cv2.threshold(img_blurred,
thresh=0,
maxval=255,
type=cv2.THRESH_BINARY_INV
| cv2.THRESH_OTSU)
# Find contours
contours, _ = cv2.findContours(img_thresh,
mode=cv2.RETR_LIST,
method=cv2.CHAIN_APPROX_SIMPLE)
# Prepare data
contours_dict = []
for contour in contours:
x, y, w, h = cv2.boundingRect(contour)
# instert to dict
contours_dict.append({
'contour': contour,
'x': x,
'y': y,
'w': w,
'h': h,
'cx': x + (w / 2),
'cy': y + (h / 2)
})
# Select candidates by char size
MIN_AREA = 80
MIN_WIDTH, MIN_HEIGHT = 2, 8
MIN_RATIO, MAX_RATIO = 0.25, 1.0
possible_contours = []
cnt = 0
for d in contours_dict:
area = d['w'] * d['h']
ratio = d['w'] / d['h']
if area > MIN_AREA \
and d['w'] > MIN_WIDTH and d['h'] > MIN_HEIGHT \
and MIN_RATIO < ratio < MAX_RATIO:
d['idx'] = cnt
cnt += 1
possible_contours.append(d)
# Select candidates by arrangement of contours
MAX_DIAG_MULTIPLYER = 5 # 5
MAX_ANGLE_DIFF = 12.0 # 12.0
MAX_AREA_DIFF = 0.5 # 0.5
MAX_WIDTH_DIFF = 0.8 # 0.8
MAX_HEIGHT_DIFF = 0.2 # 0.2
MIN_N_MATCHED = 5 # 3
def find_chars(contour_list):
matched_result_idx = []
for d1 in contour_list:
matched_contours_idx = []
for d2 in contour_list:
if d1['idx'] == d2['idx']:
continue
dx = abs(d1['cx'] - d2['cx'])
dy = abs(d1['cy'] - d2['cy'])
diagonal_length = np.sqrt(d1['w']**2 + d1['h']**2)
distance = np.linalg.norm(
np.array((d1['cx'], d1['cy'])) -
np.array((d2['cx'], d2['cy'])))
if dx == 0:
angle_diff = 90
else:
angle_diff = np.degrees(np.arctan(dy / dx))
area_diff = abs(d1['w'] * d1['h'] -
d2['w'] * d2['h']) / (d1['w'] * d1['h'])
width_diff = abs(d1['w'] - d2['w']) / d1['w']
height_diff = abs(d1['h'] - d2['h']) / d1['h']
if distance < diagonal_length * MAX_DIAG_MULTIPLYER \
and angle_diff < MAX_ANGLE_DIFF and area_diff < MAX_AREA_DIFF \
and width_diff < MAX_WIDTH_DIFF and height_diff < MAX_HEIGHT_DIFF:
matched_contours_idx.append(d2['idx'])
# append this contour
matched_contours_idx.append(d1['idx'])
if len(matched_contours_idx) < MIN_N_MATCHED:
continue
matched_result_idx.append(matched_contours_idx)
unmatched_contour_idx = []
for d4 in contour_list:
if d4['idx'] not in matched_contours_idx:
unmatched_contour_idx.append(d4['idx'])
unmatched_contour = np.take(possible_contours,
unmatched_contour_idx)
# recursive
recursive_contour_list = find_chars(unmatched_contour)
for idx in recursive_contour_list:
matched_result_idx.append(idx)
break
# optimizing
ret = []
for idx_list in matched_result_idx:
matched_contour = np.take(possible_contours, idx_list)
sorted_contour = sorted(matched_contour, key=lambda x: x['x'])
matched = []
for i in range(len(sorted_contour) - 1):
d1 = sorted_contour[i]
d2 = sorted_contour[i + 1]
if len(matched) == 0:
matched.append(d1['idx'])
diagonal_length = np.sqrt(d1['w']**2 + d1['h']**2)
distance = np.linalg.norm(
np.array((d1['cx'], d1['cy'])) -
np.array((d2['cx'], d2['cy'])))
if distance > diagonal_length * 3:
sorted_contour = sorted_contour[:i + 1]
break
matched.append(d2['idx'])
if len(matched) > 0:
ret.append(matched)
# return matched_result_idx
return ret
result_idx = find_chars(possible_contours)
matched_result = []
for idx_list in result_idx:
matched_result.append(np.take(possible_contours, idx_list))
# Rotate plate image
PLATE_WIDTH_PADDING = 1.1
# PLATE_HEIGHT_PADDING = 1.1
MIN_PLATE_RATIO = 3
MAX_PLATE_RATIO = 12
plate_imgs = []
plate_infos = []
for matched_chars in matched_result:
sorted_chars = sorted(matched_chars, key=lambda x: x['x'])
plate_cx = (sorted_chars[0]['cx'] + sorted_chars[-1]['cx']) / 2
plate_cy = (sorted_chars[0]['cy'] + sorted_chars[-1]['cy']) / 2
plate_width = (sorted_chars[-1]['x'] + sorted_chars[-1]['w'] -
sorted_chars[0]['x']) * PLATE_WIDTH_PADDING
sum_height = 0
for d in sorted_chars:
sum_height += d['h']
plate_height = int(sum_height / len(sorted_chars) *
PLATE_WIDTH_PADDING)
triangle_height = sorted_chars[-1]['cy'] - sorted_chars[0]['cy']
triangle_hypotenus = np.linalg.norm(
np.array((sorted_chars[0]['cx'], sorted_chars[0]['cy'])) -
np.array((sorted_chars[-1]['cx'], sorted_chars[-1]['cy'])))
angle = np.degrees(np.arcsin(triangle_height / triangle_hypotenus))
rotation_matrix = cv2.getRotationMatrix2D(center=(plate_cx,
plate_cy),
angle=angle,
scale=1.0)
img_rotated = cv2.warpAffine(img_thresh,
M=rotation_matrix,
dsize=(width, height))
img_cropped = cv2.getRectSubPix(img_rotated,
patchSize=(int(plate_width),
int(plate_height)),
center=(int(plate_cx),
int(plate_cy)))
ratio = img_cropped.shape[1] / img_cropped.shape[0]
if ratio < MIN_PLATE_RATIO or ratio > MAX_PLATE_RATIO:
continue
plate_imgs.append(img_cropped)
plate_infos.append({
'x': int(plate_cx - plate_width / 2),
'y': int(plate_cy - plate_height / 2),
'w': int(plate_width),
'h': int(plate_height)
})
# Another thresholding to find chars
longest_idx, longest_text = -1, 0
plate_chars = []
for i, plate_img in enumerate(plate_imgs):
plate_img = cv2.resize(plate_img, dsize=(0, 0), fx=1.6, fy=1.6)
_, plate_img = cv2.threshold(plate_img,
thresh=0.0,
maxval=255.0,
type=cv2.THRESH_BINARY
| cv2.THRESH_OTSU)
# find contours again (same as above)
contours, _ = cv2.findContours(plate_img,
mode=cv2.RETR_LIST,
method=cv2.CHAIN_APPROX_SIMPLE)
plate_min_x, plate_min_y = plate_img.shape[1], plate_img.shape[0]
plate_max_x, plate_max_y = 0, 0
for contour in contours:
x, y, w, h = cv2.boundingRect(contour)
area = w * h
ratio = w / h
if area > MIN_AREA and w > MIN_WIDTH and h > MIN_HEIGHT and MIN_RATIO < ratio < MAX_RATIO:
if x < plate_min_x:
plate_min_x = x
if y < plate_min_y:
plate_min_y = y
if x + w > plate_max_x:
plate_max_x = x + w
if y + h > plate_max_y:
plate_max_y = y + h
img_result = plate_img[plate_min_y:plate_max_y,
plate_min_x:plate_max_x]
img_result = cv2.GaussianBlur(img_result, ksize=(3, 3), sigmaX=0)
_, img_result = cv2.threshold(img_result,
thresh=0.0,
maxval=255.0,
type=cv2.THRESH_BINARY
| cv2.THRESH_OTSU)
# dilation
kernel = np.ones((2, 2), np.uint8)
img_result = cv2.dilate(img_result, kernel=kernel)
img_result = cv2.copyMakeBorder(img_result,
top=10,
bottom=10,
left=10,
right=10,
borderType=cv2.BORDER_CONSTANT,
value=(0, 0, 0))
chars = pytesseract.image_to_string(img_result,
lang='kor',
config='--psm 7 --oem 0')
result_chars = ''
has_digit = False
for c in chars:
if ord('가') <= ord(c) <= ord('힣') or c.isdigit():
if c.isdigit():
has_digit = True
result_chars += c
plate_chars.append(result_chars)
if has_digit and len(result_chars) > longest_text:
longest_idx = i
longest_text = len(result_chars)
# Result
if len(plate_chars) == 0:
gc.collect()
return None
# info = plate_infos[longest_idx]
chars = plate_chars[longest_idx]
# print(chars)
gc.collect()
return chars
isSignal = isSignal - 1
# Из сглаженного текущего кадра (t_plus) вычитаем предыдущий (t_minus).
#t_minus = cv2.GaussianBlur(t_minus, (5,5), 0)
t_plus = cv2.GaussianBlur(t_plus, (5, 5), 0)
mask = cv2.absdiff(t_minus, t_plus)
# Сравниваем значения полученной разности с некоторым пороговым значением
retval, mask = cv2.threshold(mask, 25, 255, cv2.THRESH_BINARY)
# Применяем морфологические операции закрытия и открытия, чтобы избавиться от
# движущихся регионов малого размера (шумы камеры)
kernel = np.ones((5, 5), np.uint8)
mask = cv2.morphologyEx(mask, cv2.MORPH_OPEN, kernel)
mask = cv2.morphologyEx(mask, cv2.MORPH_CLOSE, kernel)
mask_min = cv2.getRectSubPix(
mask, (width, width), (point1[0] + width / 2, point1[1] + width / 2))
countNonZero = cv2.countNonZero(mask_min)
if intervalId > 0:
intervalId = intervalId - 1
else:
intervalId = 1000
print 'countNonZero = ', countNonZero
# Если обнаружено движение
if countNonZero > countLimit:
print 'motion detected, countNonZero = ', countNonZero
# Сохранение фотки в файл
if saveToFile:
params = list()
def detect (self, image, threshold = 0.7):
match_method = cv2.TM_CCOEFF_NORMED
detected = {}
num_objects = 0
with Timer('detection'):
# Compare squares against patterns, and if not recognized, take the sample as new pattern
for i, cnt in enumerate(self.squares):
max_found = threshold
max_item = None
max_location = (-1,-1)
max_angle = -1
bounds = cv2.boundingRect(cnt)
location_abs, size, phi = cv2.minAreaRect(cnt)
cropped = cv2.getRectSubPix(self.processed, bounds[2:], location_abs)
location = (location_abs[0]-bounds[0], location_abs[1]-bounds[1])
w, h = tuple(int(c) for c in size)
rot = cv2.getRotationMatrix2D(location, phi, 1)
rotated = cv2.warpAffine(cropped, rot, dsize = cropped.shape[:2], flags=cv2.INTER_CUBIC)
sample = cv2.getRectSubPix(rotated, (w-4, h-4), location)
#sample = cv2.cvtColor(sample, cv2.COLOR_BGR2GRAY)
sample = cv2.resize(sample, (90, 90))
#ret, sample = cv2.threshold(sample, 230, 255, cv2.THRESH_BINARY)
#sample = cv2.equalizeHist(sample)
#cv2.circle(self.original, tuple(int(l) for l in location_abs), 20, (0,0,255))
#cv2.imshow('crop'+str(i),self.processed)
for tag, pattern in self.patterns.items():
match = cv2.matchTemplate(sample, pattern, match_method)
result, _, _, _ = cv2.minMaxLoc(match)
if result > max_found:
#print "Best candidate for %s is %s with %s" % (str(i),tag, str(result))
max_found = result
max_item = tag
max_location = (int(location_abs[0]), int(location_abs[1]))
max_angle = phi
if result > threshold * 1.5:
break
else:
#print "Discarted", str(i)
if self.training:
#print "Store new pattern",i
cv2.imwrite('patterns/tag-%s.pgm' % str(i), sample)
if max_item:
cv2.drawContours(self.original, self.squares, i, (0, 255, 0), 5)
tag_id = max_item.replace("patterns/tag-", "").replace(".pgm", "")
max_angle = (360 - max_angle) % 360
try:
name, angle = tag_id.split("-", 1)
#if name=='arrow':
# print "Orig angle", max_angle, "using", angle
if angle=='t1':
max_angle += 90
elif angle=='t2':
max_angle += 180
elif angle=='t3':
max_angle += 270
except:
name = tag_id
cv2.putText(self.original, "%s" % name,
(max_location[0]-20, max_location[1]-20), cv2.FONT_HERSHEY_PLAIN, 1, (0,0,255), 1)
#vmod = 10
#p2 = (int(max_location[0] + vmod * np.cos(max_angle + np.pi/4)),
# int(max_location[1] + vmod * np.sin(max_angle + np.pi/4)))
#cv2.line(self.original, max_location, p2, (0,255,255), 3)
#print "Chosen %s to be %s" % (str(i), max_item)
if not name in detected:
detected[name] = []
detected[name].append({'angle': int(100 * max_angle)/100.0, 'pos': max_location})
num_objects +=1
return num_objects, detected
cv2.CHAIN_APPROX_NONE)
contours_after_size_verification = []
for contour in contours:
if verifySize(height_img, contour):
contours_after_size_verification.append(contour)
for contour in contours_after_size_verification:
# bounding rect
x, y, w, h = cv2.boundingRect(contour)
paddingw = int(w / 3)
paddingh = int(h / 4)
#img_crop used to reference copy of input, and the written file was img crop
img_crop = cv2.getRectSubPix(img_contours,
(w + paddingw, h + paddingh),
(x + w / 2, y + h / 2))
resized_image = cv2.resize(img_crop, (34, 85))
if "image" not in trace:
# file_name = trace[:-6]+'_'+str(numImg)+'.png'
file_name = trace[:-4] + '_' + str(numImg) + '.png'
elif "image" in trace:
# file_name = 'lp_' + trace[:-4] + '_' + str(numImg) + '.png'
file_name = trace[:-4] + '_' + str(numImg) + '.png'
else:
print("image not being segmented: ", trace)
break
cv2.imwrite("{}/{}".format(PATH_WRITE, file_name), resized_image)
def CarPlate(self):
global InNum, OutNum, CarNum, InTime, TotalTime, InTimeDB, TotalFee
plt.style.use('dark_background')
############차 이미지 가져오기
img_car = cv2.imread('car.jpg')
height, width, channel = img_car.shape # 사진 크기
############그레이스케일 변환
gray = cv2.cvtColor(img_car, cv2.COLOR_BGR2GRAY)
############Maximize Contrast 대비
structuringElement = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
imgTopHat = cv2.morphologyEx(
gray, cv2.MORPH_TOPHAT, structuringElement) # Opeining과 원본 이미지의 차이
imgBlackHat = cv2.morphologyEx(
gray, cv2.MORPH_BLACKHAT,
structuringElement) # Closing과 원본 이미지의 차이
imgGrayscalePlusTopHat = cv2.add(gray, imgTopHat)
gray = cv2.subtract(imgGrayscalePlusTopHat, imgBlackHat)
###########노이즈 제거 블러 > thereshold
img_blurred = cv2.GaussianBlur(gray, ksize=(5, 5), sigmaX=0)
img_thresh = cv2.adaptiveThreshold(
img_blurred,
maxValue=255.0,
adaptiveMethod=cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
thresholdType=cv2.THRESH_BINARY_INV,
blockSize=19,
C=9)
# 윤곽선 검출 temp_result
contours, _ = cv2.findContours(img_thresh,
mode=cv2.RETR_LIST,
method=cv2.CHAIN_APPROX_SIMPLE)
#####윤곽선을 감싸는 사각형 그리기
temp_result = np.zeros((height, width, channel), dtype=np.uint8)
contours_dict = []
for contour in contours:
x, y, w, h = cv2.boundingRect(contour)
cv2.rectangle(temp_result,
pt1=(x, y),
pt2=(x + w, y + h),
color=(255, 255, 255),
thickness=2)
# insert to dict
contours_dict.append({
'contour': contour,
'x': x,
'y': y,
'w': w,
'h': h,
'cx': x + (w / 2),
'cy': y + (h / 2)
})
# 번호판 후보 찾기
MIN_AREA = 80
MIN_WIDTH, MIN_HEIGHT = 2, 8
MIN_RATIO, MAX_RATIO = 0.25, 1.0
possible_contours = []
cnt = 0
for d in contours_dict:
area = d['w'] * d['h']
ratio = d['w'] / d['h']
if area > MIN_AREA \
and d['w'] > MIN_WIDTH and d['h'] > MIN_HEIGHT \
and MIN_RATIO < ratio < MAX_RATIO:
d['idx'] = cnt
cnt += 1
possible_contours.append(d)
# 윤곽선의 배열로 최종 후보 선정
MAX_DIAG_MULTIPLYER = 5 # 번호사이 간격
MAX_ANGLE_DIFF = 12.0 # 번호 사이 각도
MAX_AREA_DIFF = 0.5 # 번호 면적 차이
MAX_WIDTH_DIFF = 0.8 # 너비차이
MAX_HEIGHT_DIFF = 0.2 # 높이차이
MIN_N_MATCHED = 3 # 위의 5조건 만족하는 최소 개수
# recursive 방식으로 찾기
def find_chars(contour_list):
matched_result_idx = [] # 최종적으로 남는 후보의 인덱스 저장
for d1 in contour_list:
matched_contours_idx = []
for d2 in contour_list:
if d1['idx'] == d2['idx']:
continue
dx = abs(d1['cx'] - d2['cx'])
dy = abs(d1['cy'] - d2['cy'])
#
diagonal_length1 = np.sqrt(d1['w']**2 + d1['h']**2)
# 윤곽선 중심과 중심 사이의 거리 구하기
distance = np.linalg.norm(
np.array([d1['cx'], d1['cy']]) -
np.array([d2['cx'], d2['cy']]))
if dx == 0:
angle_diff = 90 ##윤곽선 사이 각
else:
angle_diff = np.degrees(np.arctan(dy / dx))
area_diff = abs(d1['w'] * d1['h'] -
d2['w'] * d2['h']) / (d1['w'] * d1['h'])
width_diff = abs(d1['w'] - d2['w']) / d1['w']
height_diff = abs(d1['h'] - d2['h']) / d1['h']
if distance < diagonal_length1 * MAX_DIAG_MULTIPLYER \
and angle_diff < MAX_ANGLE_DIFF and area_diff < MAX_AREA_DIFF \
and width_diff < MAX_WIDTH_DIFF and height_diff < MAX_HEIGHT_DIFF:
matched_contours_idx.append(d2['idx'])
# append this contour
matched_contours_idx.append(d1['idx'])
if len(matched_contours_idx) < MIN_N_MATCHED:
continue
matched_result_idx.append(matched_contours_idx)
unmatched_contour_idx = []
for d4 in contour_list:
if d4['idx'] not in matched_contours_idx:
unmatched_contour_idx.append(d4['idx'])
unmatched_contour = np.take(
possible_contours, unmatched_contour_idx) # 인덱스 같은 값만 추출
# recursive
recursive_contour_list = find_chars(unmatched_contour)
for idx in recursive_contour_list:
matched_result_idx.append(idx)
break
return matched_result_idx
result_idx = find_chars(possible_contours)
matched_result = []
for idx_list in result_idx:
matched_result.append(np.take(possible_contours, idx_list))
# 기울어진 이미지 회전
PLATE_WIDTH_PADDING = 1.3 # 1.3
PLATE_HEIGHT_PADDING = 1.5 # 1.5
MIN_PLATE_RATIO = 3
MAX_PLATE_RATIO = 10
plate_imgs = []
plate_infos = []
for i, matched_chars in enumerate(matched_result):
sorted_chars = sorted(matched_chars, key=lambda x: x['cx'])
plate_cx = (sorted_chars[0]['cx'] + sorted_chars[-1]['cx']) / 2
plate_cy = (sorted_chars[0]['cy'] + sorted_chars[-1]['cy']) / 2
plate_width = (sorted_chars[-1]['x'] + sorted_chars[-1]['w'] -
sorted_chars[0]['x']) * PLATE_WIDTH_PADDING
sum_height = 0
for d in sorted_chars:
sum_height += d['h']
plate_height = int(sum_height / len(sorted_chars) *
PLATE_HEIGHT_PADDING)
triangle_height = sorted_chars[-1]['cy'] - sorted_chars[0]['cy']
triangle_hypotenus = np.linalg.norm(
np.array([sorted_chars[0]['cx'], sorted_chars[0]['cy']]) -
np.array([sorted_chars[-1]['cx'], sorted_chars[-1]['cy']]))
angle = np.degrees(np.arcsin(triangle_height / triangle_hypotenus))
rotation_matrix = cv2.getRotationMatrix2D(center=(plate_cx,
plate_cy),
angle=angle,
scale=1.0)
img_rotated = cv2.warpAffine(img_thresh,
M=rotation_matrix,
dsize=(width, height))
img_cropped = cv2.getRectSubPix(img_rotated,
patchSize=(int(plate_width),
int(plate_height)),
center=(int(plate_cx),
int(plate_cy)))
if img_cropped.shape[1] / img_cropped.shape[0] < MIN_PLATE_RATIO or img_cropped.shape[1] / \
img_cropped.shape[
0] < MIN_PLATE_RATIO > MAX_PLATE_RATIO:
continue
plate_imgs.append(img_cropped)
plate_infos.append({
'x': int(plate_cx - plate_width / 2),
'y': int(plate_cy - plate_height / 2),
'w': int(plate_width),
'h': int(plate_height)
})
####한번더 threshold
longest_idx, longest_text = -1, 0
plate_chars = []
for i, plate_img in enumerate(plate_imgs):
plate_img = cv2.resize(plate_img, dsize=(0, 0), fx=1.6, fy=1.6)
_, plate_img = cv2.threshold(plate_img,
thresh=0.0,
maxval=255.0,
type=cv2.THRESH_BINARY
| cv2.THRESH_OTSU)
contours, _ = cv2.findContours(plate_img,
mode=cv2.RETR_LIST,
method=cv2.CHAIN_APPROX_SIMPLE)
plate_min_x, plate_min_y = plate_img.shape[1], plate_img.shape[0]
plate_max_x, plate_max_y = 0, 0
for contour in contours:
x, y, w, h = cv2.boundingRect(contour)
area = w * h
ratio = w / h
if area > MIN_AREA \
and w > MIN_WIDTH and h > MIN_HEIGHT \
and MIN_RATIO < ratio < MAX_RATIO:
if x < plate_min_x:
plate_min_x = x
if y < plate_min_y:
plate_min_y = y
if x + w > plate_max_x:
plate_max_x = x + w
if y + h > plate_max_y:
plate_max_y = y + h
img_result = plate_img[plate_min_y:plate_max_y,
plate_min_x:plate_max_x]
img_result = cv2.GaussianBlur(img_result, ksize=(3, 3), sigmaX=0)
_, img_result = cv2.threshold(img_result,
thresh=0.0,
maxval=255.0,
type=cv2.THRESH_BINARY
| cv2.THRESH_OTSU)
img_result = cv2.copyMakeBorder(img_result,
top=10,
bottom=10,
left=10,
right=10,
borderType=cv2.BORDER_CONSTANT,
value=(0, 0, 0))
####글자 인식
text = pytesseract.image_to_string(img_result,
lang='kor',
config='--psm 7 --oem 0')
result_chars = ''
has_digit = False ##숫자포함문자열
for c in text:
if ord('가') <= ord(c) <= ord('힣') or c.isdigit():
if c.isdigit():
has_digit = True
result_chars += c
plate_chars.append(result_chars)
if has_digit and len(result_chars) > longest_text:
longest_idx = i
info = plate_infos[longest_idx]
CarNum = plate_chars[longest_idx]
print("차량 번호 ", CarNum)
img_out = img_car.copy()
cv2.rectangle(img_out,
pt1=(info['x'], info['y']),
pt2=(info['x'] + info['w'], info['y'] + info['h']),
color=(255, 0, 0),
thickness=2)
CarNUmber = plt.figure(figsize=(8, 6))
plt.imshow(img_out)
plt.show()
reply = QMessageBox.question(self, '차량 번호 확인',
'차량 번호는 %s가 맞습니까?' % CarNum,
QMessageBox.Yes | QMessageBox.No,
QMessageBox.No)
if reply == QMessageBox.Yes:
plt.close()
cv2.destroyWindow('Car Video')
else:
exit()
def get_im(self, src_im, rect=(0, 0, 100, 100)):
return cv2.getRectSubPix(src_im,
(rect[2] - rect[0], rect[3] - rect[1]),
((rect[2] + rect[0]) / 2,
(rect[3] + rect[1]) / 2))
import cv2
import os
import glob
img_dir = "C:/users/hp/image/" # Enter Directory of all images
data_path = os.path.join(img_dir, '*g')
files = glob.glob(data_path)
for f1 in files:
img = cv2.imread(f1)
img = cv2.getRectSubPix(img, (320, 220), (150, 170))
cv2.imwrite(f1, img)