def play(filepath):
global img
global rects
global video
cv2.namedWindow(win_name)
cv2.cv.SetMouseCallback(win_name, on_mouse, 0)
clf = model.load_classifier()
pos = 0 # frame position
video = cv2.VideoCapture(filepath)
frames = video.get(cv2.cv.CV_CAP_PROP_FRAME_COUNT)
cv2.cv.CreateTrackbar('tracker', win_name, pos, int(frames), on_change)
key_val = 0
while(video.isOpened()):
ret, img = video.read()
if not ret:
break
if key_val == 0:
cv2.imshow(win_name, img)
pos = pos + 1
c = cv2.waitKey(key_val)
if c == ord('c'):
key_val = 1
cv2.cv.SetMouseCallback(win_name, on_mouse_do_nothing, 0)
elif c == ord('q'):
break
if key_val==1:
if op.has_score_board(img, target_imgs[0], rects[0]):
show_recognize(clf, img, rects[1:])
cv2.imshow(win_name, img)
video.release()
cv2.destroyAllWindows()
def play(filepath):
global img
global rects
global video
cv2.namedWindow(win_name)
cv2.cv.SetMouseCallback(win_name, on_mouse, 0)
clf = model.load_classifier()
pos = 0 # frame position
video = cv2.VideoCapture(filepath)
frames = video.get(cv2.cv.CV_CAP_PROP_FRAME_COUNT)
cv2.cv.CreateTrackbar('tracker', win_name, pos, int(frames), on_change)
key_val = 0
while (video.isOpened()):
ret, img = video.read()
if not ret:
break
if key_val == 0:
cv2.imshow(win_name, img)
pos = pos + 1
c = cv2.waitKey(key_val)
if c == ord('c'):
key_val = 1
cv2.cv.SetMouseCallback(win_name, on_mouse_do_nothing, 0)
elif c == ord('q'):
break
if key_val == 1:
if op.has_score_board(img, target_imgs[0], rects[0]):
show_recognize(clf, img, rects[1:])
cv2.imshow(win_name, img)
video.release()
cv2.destroyAllWindows()
def get_classifier(modelfile, cache=True, use_classify_many=False):
classifier, encoder = load_classifier(modelfile)
cache_dict = {}
classes = encoder.classes_
def classify(name):
name = name.upper()
if name in cache_dict:
return cache_dict[name]
else:
feature = gen_features(name)
label_numeric = classifier.predict(feature)[0]
label = classes[label_numeric]
if cache:
cache_dict[name] = label
return label
classify.classifier = classifier
def classify_many(names):
features = [gen_features(name) for name in names]
labels = classifier.predict(features)
return [classes[l] for l in labels]
classify_many.classifier = classifier
if use_classify_many:
return classify_many
else:
return classify
def test_get_text(img):
imgs, pos = get_text(img)
print pos
clf = model.load_classifier()
for i,img in enumerate(imgs):
cv2.imshow(str(i),img)
print model.pred(clf,img)
cv2.waitKey(0)
cv2.destroyAllWindows()
def test_get_text(img):
imgs, pos = get_text(img)
print pos
clf = model.load_classifier()
for i, img in enumerate(imgs):
cv2.imshow(str(i), img)
print model.pred(clf, img)
cv2.waitKey(0)
cv2.destroyAllWindows()
def find_num_pos(video, lines):
fps = video.get(cv2.cv.CV_CAP_PROP_FPS)
if len(lines) == 0:
print 'no lines'
return
video.set(cv2.cv.CV_CAP_PROP_POS_FRAMES, 0)
clf = model.load_classifier()
video.set(cv2.cv.CV_CAP_PROP_POS_FRAMES, fps * 60 * 15)
while (video.isOpened):
ret, cur = video.read()
if not ret:
break
key = cv2.waitKey(1)
if key == ord('q'):
break
for line in lines:
imgs, pos = nc.get_text(cur[line[0] + 1:line[1] - 1])
# cv2.imshow('binary', board)
# print len(imgs)
for i, img in enumerate(imgs):
num, prob = model.pred_prob(clf, img)
if prob > 0.2:
continue
cv2.putText(cur,
str(int(num)), (pos[i][0], line[0]),
cv2.FONT_HERSHEY_SIMPLEX,
0.6, (0, 0, 0xff),
thickness=2)
cv2.imshow('cur', cur)
# binary = op.get_binary(cur)
# edges = cv2.Canny(binary,50,150,apertureSize = 3)
# lines = cv2.HoughLines(edges,1,np.pi/180,200)
# print edges, lines
# if lines != None:
video.release()
cv2.destroyAllWindows()
def find_num_pos(video, lines):
fps = video.get(cv2.cv.CV_CAP_PROP_FPS)
if len(lines) == 0:
print 'no lines'
return
video.set(cv2.cv.CV_CAP_PROP_POS_FRAMES, 0)
clf = model.load_classifier()
video.set(cv2.cv.CV_CAP_PROP_POS_FRAMES, fps * 60 * 15)
while(video.isOpened):
ret, cur = video.read()
if not ret:
break
key = cv2.waitKey(1)
if key == ord('q'):
break
for line in lines:
imgs, pos = nc.get_text(cur[line[0]+1:line[1]-1])
# cv2.imshow('binary', board)
# print len(imgs)
for i, img in enumerate(imgs):
num, prob = model.pred_prob(clf, img)
if prob > 0.2:
continue
cv2.putText(cur, str(int(num)),(pos[i][0],line[0]), cv2.FONT_HERSHEY_SIMPLEX,0.6, (0,0,0xff), thickness = 2)
cv2.imshow('cur', cur)
# binary = op.get_binary(cur)
# edges = cv2.Canny(binary,50,150,apertureSize = 3)
# lines = cv2.HoughLines(edges,1,np.pi/180,200)
# print edges, lines
# if lines != None:
video.release()
cv2.destroyAllWindows()
def get_score_info(video):
global step
score_frames = []
twenty_four_frames = []
camera_change_frames = []
score_a, score_b = 0, 0 # score of both team
last_sec = 0 # 24 second board
# last_frame = None
pos = 0 # frame position
video.set(cv2.cv.CV_CAP_PROP_POS_FRAMES, pos)
clf = model.load_classifier()
while (video.isOpened()):
if pos % step == 0:
ret, frame = video.red()
else:
ret = video.grab()
if not ret:
break
# skip each step steps
if pos % step != 0:
pos += 1
continue
# if camera change
if pos != 0 and hist.camera_change(last_frame, frame):
camera_change_frames.append(pos)
print 'camera change', pos
last_frame = frame
if op.has_score_board(frame, target_imgs[0], rects[0]):
# score a
img = op.get_subimage(frame, rects[1][0], rects[1][1])
num = model.recognize_number(clf, img)
# print 'a ', num
if score_two_or_three(num, score_a):
score_frames.append(pos)
print 'a from', score_a, 'to', num, pos
if not noise(num, score_a):
score_a = num
# score b
img = op.get_subimage(frame, rects[2][0], rects[2][1])
num = model.recognize_number(clf, img)
# print 'b ', num
if score_two_or_three(num, score_b):
score_frames.append(pos)
print 'b from', score_b, 'to', num, pos
if not noise(num, score_b):
score_b = num
# 24 second board
img = op.get_subimage(frame, rects[3][0], rects[3][1])
sec = model.recognize_number(clf, img)
if last_sec != 24 and sec == 24:
print 'twenty four', pos
twenty_four_frames.append(pos)
last_sec = sec
pos += 1
# print pos
print score_frames, twenty_four_frames, camera_change_frames
return score_frames, twenty_four_frames, camera_change_frames
def get_score_info(video):
global step
score_frames = []
twenty_four_frames = []
camera_change_frames = []
score_a, score_b = 0,0 # score of both team
last_sec = 0 # 24 second board
# last_frame = None
pos = 0 # frame position
video.set(cv2.cv.CV_CAP_PROP_POS_FRAMES, pos)
clf = model.load_classifier()
while(video.isOpened()):
if pos % step == 0:
ret, frame = video.red()
else:
ret = video.grab()
if not ret:
break
# skip each step steps
if pos % step != 0:
pos += 1
continue
# if camera change
if pos != 0 and hist.camera_change(last_frame, frame):
camera_change_frames.append(pos)
print 'camera change', pos
last_frame = frame
if op.has_score_board(frame, target_imgs[0], rects[0]):
# score a
img = op.get_subimage(frame, rects[1][0], rects[1][1])
num = model.recognize_number(clf, img)
# print 'a ', num
if score_two_or_three(num,score_a):
score_frames.append(pos)
print 'a from',score_a,'to',num, pos
if not noise(num, score_a):
score_a = num
# score b
img = op.get_subimage(frame, rects[2][0], rects[2][1])
num = model.recognize_number(clf, img)
# print 'b ', num
if score_two_or_three(num,score_b):
score_frames.append(pos)
print 'b from',score_b,'to', num, pos
if not noise(num, score_b):
score_b = num
# 24 second board
img = op.get_subimage(frame, rects[3][0], rects[3][1])
sec = model.recognize_number(clf, img)
if last_sec != 24 and sec == 24:
print 'twenty four',pos
twenty_four_frames.append(pos)
last_sec = sec
pos += 1
# print pos
print score_frames, twenty_four_frames, camera_change_frames
return score_frames,twenty_four_frames,camera_change_frames
