def detect_by_gf(origin,
isdebug=False):
if origin is None:
return None
# Default Size
h,w,c = origin.shape
size = 200.0
# Resize
img = cv2.resize(origin,(int(w*size/h),int(size)))
# Extract Good Features
corners = refinedGoodFeatures(origin,img,
model='LP')
mask = checkFeatures(img,corners,True)
# Find Candidate
bboxes = findBBox(img,mask,
model='LP',
debug=True)
# Resize
if bboxes is not None:
bboxes = resizeBBoxes(bboxes,h/size)
# Check Result
if isdebug and bboxes is not None:
drawBBox(origin,bboxes,debug=True)
# Crop Rois
rois = BBoxes2ROIs(origin,bboxes)
if isdebug and rois is not None:
for i in range(len(rois)):
showResult("cropped",rois[i])
return bboxes,rois
def processLP(image, bbox_car, bboxes_lp, confidence):
markImg = None
if bboxes_lp is not None:
# Select LP with highest confidence
bbox_lp = bboxes_lp[0]
###
if confidence > 0.5:
app.results.append(r"车牌 : 有")
# Mark Image
bbox_lp_refined = shiftBBoxes(
bbox_car, [bbox_lp]) if bbox_car is not None else bbox_lp
markImg = drawBBox(image, [bbox_lp_refined], bbox_car)
#
if confidence > 0.85 and bbox_car is not None:
app.results.append(r"车牌 : 全面")
res, reps = detect_angle(image, bbox_lp_refined, bbox_car)
app.results.append(r"分析结果 : " + res)
for rep in reps:
app.results.append(rep)
else:
app.results.append(r"分析结果 : 没通过")
app.results.append(r"车牌 : 不全面")
else:
app.results.append(r"车牌 : 没有")
markImg = drawBBox(image, None, bbox_car)
else:
app.results.append(r"车牌 : 不全面")
markImg = drawBBox(image, None, bbox_car)
return markImg
def processLP(image,bbox_car,bboxes_lp,confidence):
markImg = None
if bboxes_lp is not None:
# Select LP with highest confidence
bbox_lp = bboxes_lp[0]
###
if confidence > 0.5:
app.results.append(r"License plate : Yes")
# Mark Image
bbox_lp_refined = shiftBBoxes(bbox_car,[bbox_lp]) if bbox_car is not None else bbox_lp
markImg = drawBBox(image,[bbox_lp_refined],bbox_car)
#
if confidence > 0.85 and bbox_car is not None:
app.results.append(r"License plate : FRONT")
res,reps = detect_angle(image,bbox_lp_refined,bbox_car)
app.results.append(r"Analysis result : " + res)
for rep in reps:
app.results.append(rep)
else:
app.results.append(r"Analysis result : No Pass " + str(confidence))
app.results.append(r"License plate : No Front")
else:
app.results.append(r"License plate : NO")
markImg = drawBBox(image,None,bbox_car)
else:
app.results.append(r"License plate : No Front")
markImg = drawBBox(image,None,bbox_car)
return markImg
def detect_by_data(self, img, debug=False):
# Detect
bbox = self.detect(img) #mpimg.imread(path)
# Check Result
if bbox is not None and debug:
drawBBox(img, [bbox])
##
return bbox
def detect(img, debug=False):
rclasses, rscores, rbboxes = process_image(img)
#visualization.plt_bboxes(img, rclasses, rscores, rbboxes)
# Refine BBoxes
bbox = VehicleDetector.pick_one_vehicle(img, rclasses, rscores, rbboxes)
if debug:
drawBBox(img, [bbox], drawmline=False)
return bbox
def detect_by_filename(self, path, debug=False):
# Load File
img = mpimg.imread(path) #img = cv2.imread(path)
# Detect
bbox = self.detect(img) #mpimg.imread(path)
# Check Result
if bbox is not None and debug:
drawBBox(img, [bbox])
##
return bbox
def detect_by_seg_gf(origin,
isdebug=False):
if origin is None:
return None
# Default Size
h,w,c = origin.shape
size = 200.0
#origin = opencv2skimage(origin)
# Resize
img = cv2.resize(origin,(int(w*size/h),int(size)))
# Blur
blur = cv2.GaussianBlur(img,(5,5),3)
# Equalization Hist
#origin = equalizehist(origin)
# Extract Good Features
corners = refinedGoodFeatures(origin,img,
model='LP')
corners_,handwrite = refinedCorners(img,corners,False)
checkFeatures(img,corners,True)
# Opencv2Skimage
skimg = cv2.cvtColor(blur, cv2.COLOR_BGR2RGB)
# Segmentation
out,labels = seg(skimg,Debug=False)
# Eval Label
labels = refineLabels(out,labels,corners_,howmany=20)
# Show Result
out = drawLabels(skimg,labels,Debug=isdebug)
if out is None:
return None
changebgcolr(out,labels)
#showResult("labelout",skimage2opencv(out))
# Find Candidate
bboxes = findBBox(img,out,
model='LP',
debug=isdebug)
# Resize
if bboxes is not None:
bboxes = resizeBBoxes(bboxes,h/size)
# Check Candidate
if isdebug and bboxes is not None:
drawBBox(origin,bboxes,debug=isdebug)
# Crop Rois
rois = BBoxes2ROIs(origin,bboxes)
if isdebug and rois is not None:
for i in range(len(rois)):
showResult("cropped",rois[i])
'''
if isdebug:
#labels2boundaries(labels)
contours = labels2contours(labels)
drawBBox(img,contours,debug=True)
'''
return bboxes,rois
def detect_by_probability(origin,
isdebug=False):
if origin is None:
return None
# Default Size
h,w,c = origin.shape
size = 200.0
# Resize
img = cv2.resize(origin,(int(w*size/h),int(size)))
#showResult("img",img)
#
if dayornight(img):
# Extract Good Features
corners = refinedGoodFeatures(origin,img)
mask = checkFeatures(img,corners,False)
closing=close(mask)
refined_gfmask = refine_gfimage(img,closing)
#showResult("refined_gfmask",refined_gfmask)
finalmask = mkfinalmask(img,refined_gfmask,isday=True)
else:
finalmask = mkfinalmask(img,None,isday=False)
#
#showResult("masktest",finalmask)
#ret,binary = cv2.threshold(finalmask,0,255,cv2.THRESH_BINARY+cv2.THRESH_OTSU)
binary = compositeThreshold(finalmask,mode='otsu')
kernel = cv2.getStructuringElement(cv2.MORPH_RECT,(5,5))
closing=cv2.morphologyEx(binary, cv2.MORPH_CLOSE, kernel)
if isdebug:
showResult("masktest",closing)
# Find Candidate
bboxes = findBBox(img,closing,isdebug=True)
# Resize
if bboxes is not None:
bboxes = resizeBBoxes(bboxes,h/size)
# Check Result
if isdebug and bboxes is not None:
drawBBox(origin,bboxes,debug=True)
# Crop Rois
rois = BBoxes2ROIs(origin,bboxes)
if isdebug and rois is not None:
for i in range(len(rois)):
showResult("cropped",rois[i])
return bboxes,rois
def detect_by_cascade(origin, resize_h=720, en_scale=1.06, isdebug=False):
if origin is None:
return None
# Default Size
height, width, channels = origin.shape
# Resize
resized = cv2.resize(origin, (int(width * resize_h / height), resize_h))
# Colr2Gray
gray = cv2.cvtColor(resized, cv2.COLOR_RGB2GRAY)
# Detect by Cascade
watches = watch_cascade.detectMultiScale(gray,
en_scale,
1,
minSize=(36, 9))
cropped_images = []
bboxes = []
for (x, y, w, h) in watches:
xmin = x - w * 0.1
ymin = y - h * 0.6
xmin = xmin if xmin >= 0 else 0
xmin = ymin if ymin >= 0 else 0
xmax = xmin + 1.2 * w
ymax = ymin + 1.1 * h
bboxes.append(
np.array([[xmin, ymin], [xmax, ymin], [xmax, ymax], [xmin, ymax]]))
cropped = cropped_from_image(
gray, (int(xmin), int(ymin), int(1.2 * w), int(1.1 * h)))
cropped_images.append(cropped)
# Resize
if bboxes is not None:
bboxes = resizeBBoxes(bboxes, height / float(resize_h))
# Check Result
if isdebug and bboxes is not None:
drawBBox(origin, bboxes, debug=True)
# Crop Rois
rois = BBoxes2ROIs(origin, bboxes)
if isdebug and rois is not None:
for i in range(len(rois)):
showResult("cropped", rois[i])
return bboxes, rois #cropped_images
def main():
start = time.time()
# Load Image
image = cv2.imread(fullpath)
# Load Model
#detector = VehicleDetector()
# Detect Vehicle
bbox_car = detect(opencv2skimage(image)) #mpimg.imread(path)
if bbox_car is not None:
img_car = cropImg_by_BBox(image, bbox_car)
# Detect License Plate
start_detect_lp = time.time()
confidence, bboxes_lp, rois = lpdetector.process(img_car)
print("detecting LP elapsed time: " +
str(int((time.time() - start_detect_lp) * 1000) / 1000.0) + "s")
# Check Result
if bboxes_lp is not None:
bboxes_lp_refined = shiftBBoxes(bbox_car, bboxes_lp)
print "confidence:", confidence
drawBBox(image, bboxes_lp_refined, bbox_car, debug=True)
'''
for roi in rois:
start_refine_lp = time.time()
lp.process(roi,isdebug=True)
print("refining LP elapsed time: "+str(int((time.time() - start_refine_lp)*1000)/1000.0)+"s")
#pipline.recognizeLP2(seg_blocks,mid)
print("confidence:"+str(lp.confidence*100)+"%")
'''
else:
drawBBox(image, None, bbox_car, debug=True)
else:
start_detect_lp = time.time()
confidence, bboxes_lp, rois = lpdetector.process(image)
print("detecting LP elapsed time: " +
str(int((time.time() - start_detect_lp) * 1000) / 1000.0) + "s")
if bboxes_lp is not None:
print "confidence:", confidence
drawBBox(image, bboxes_lp, None, debug=True)
print("total elapsed time: " +
str(int((time.time() - start) * 1000) / 1000.0) + "s")
