def run(self):
while not self.video_metas.empty():
video_meta = None
try:
print 'undeal %d' % (self.video_metas.qsize(), )
video_meta = self.video_metas.get()
if not os.path.exists(video_meta.frames_file_dir):
os.makedirs(video_meta.frames_file_dir)
if os.path.exists(video_meta.meta_path) and os.path.isfile(
video_meta.meta_path):
print('video %s had deal' % (video_meta.video_path, ))
# selvideo_metas.remove(video_meta)
continue
video_shape = get_video_shape(video_meta.video_path)
frames_dirs = self.extract_frame(
video_meta,
video_shape,
video_meta.frames_file_pattern,
frame_rate=self.frame_rate,
distinct=self.distinct)
with open(video_meta.meta_path, 'wb') as meta_file:
meta_file.writelines(frames_dirs)
self.save_to_database(video_meta.meta_path)
except Exception, e:
print 'threading error'
print os.path.exists(video_meta.meta_path)
if os.path.exists(video_meta.meta_path):
os.remove(video_meta.meta_path)
raise RuntimeError()
def extract_frame(args, video_path):
video_path = os.path.abspath(video_path)
video_file = os.path.basename(video_path)
video_name = os.path.splitext(video_file)[0]
frame_rate = args.frame_rate if args.frame_rate is not None else None
frame_nums = args.frame_nums if args.frame_nums is not None and len(
args.frame_nums) > 0 else None
distinct = args.distinct if args.distinct is not None else False
video_shape = get_video_shape(video_path)
if frame_nums is None:
frame_generator = VideoFrameGenerator(video_path, video_shape,
frame_rate)
with frame_generator as frames:
i = 1
prev_frame = None
for frame in frames:
should_write = True
if prev_frame is not None and args.distinct:
should_write = not numpy.array_equal(prev_frame, frame)
if should_write:
process(args, video_path, frame, i)
prev_frame = frame
i += 1
else:
frame_nums = sorted(frame_nums)
max_frame_num = max(frame_nums)
frame_generator = VideoFrameGenerator(video_path, video_shape,
frame_rate)
with frame_generator as frames:
frame_num_index = 0
frame_num = frame_nums[frame_num_index]
i = 1
for frame in frames:
should_write = False
if frame_nums is None:
# no frame_nums, write all
should_write = True
else:
if i > max_frame_num:
break
if i == frame_num:
should_write = True
frame_num_index += 1
if frame_num_index < len(frame_nums):
frame_num = frame_nums[frame_num_index]
if should_write:
process(args, frame, i)
i += 1
def main():
parser = argparse.ArgumentParser()
parser.add_argument('-v',
'--verbose',
action='store_true',
default=False,
help='verbose')
parser.add_argument('-r', '--frame-rate', type=float, help='frame rate')
parser.add_argument('-d',
'--distinct',
action='store_true',
default=False,
help='distinct')
parser.add_argument('-n',
'--frame-nums',
nargs='*',
type=int,
help='frame nums')
parser.add_argument('-t', '--frame-type', help='frame type')
parser.add_argument('--output-dir', help='output dir')
parser.add_argument('video_paths', nargs='+')
args = parser.parse_args()
frame_rate = args.frame_rate
distinct = args.distinct
frame_nums = args.frame_nums if args.frame_nums is not None and len(
args.frame_nums) > 0 else None
frame_type = args.frame_type if args.frame_type is not None else 'jpg'
for video_path in args.video_paths:
video_path = os.path.abspath(video_path)
video_file = os.path.basename(video_path)
video_name = os.path.splitext(video_file)[0]
frames_dir = args.output_dir
if frames_dir is None:
frames_basedir = os.path.dirname(video_path)
frames_dir = os.path.join(frames_basedir, video_name)
if not os.path.exists(frames_dir):
os.mkdir(frames_dir)
video_shape = get_video_shape(video_path)
extract_frame(video_path,
video_shape,
frames_dir,
frame_rate=frame_rate,
distinct=distinct,
frame_nums=frame_nums,
frame_type=frame_type)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('-v',
'--verbose',
action='store_true',
default=False,
help='verbose')
parser.add_argument('video_path', help='video path')
parser.add_argument('frame_rate', type=float, help='frame rate')
parser.add_argument('frame_nums', nargs='+', type=int, help='frame nums')
args = parser.parse_args()
video_path = os.path.abspath(args.video_path)
frame_rate = args.frame_rate
frame_nums = args.frame_nums
video_shape = get_video_shape(video_path)
view_frame(video_path, video_shape, frame_rate, frame_nums)
def get_frame_generator(video_path, frame_rate):
video_shape = get_video_shape(video_path)
return VideoFrameGenerator(video_path, video_shape, frame_rate)
def process_image_handler(error_csv_output_path, fr_csv_output_path,
video_file_path, image_basedir, roi_setting,
settings):
frame_type = 'bmp'
video_name = os.path.basename(video_file_path).split('.')[0]
platform = video_name.split('_')[1]
app_name = video_name.split('_')[0]
# TODO 把参数获取方式整理好，现在好恶心
pre_roi_frames_dict = None
try:
pre_roi_frames_dict = settings.get_specific_settings(
platform, app_name, 'pre_roi')
except ValueError as e:
logger.info('%s_%s no need pre roi' % (app_name, platform))
resize_frames_dict = settings.get_specific_settings(
platform, app_name, 'resize')
video_shape = get_video_shape(video_file_path)
# frame_rate = get_frame_rate(video_file_path)
frame_rate = 60
if not os.path.exists(image_basedir):
os.makedirs(image_basedir)
src_align_frames_dict = settings.get_specific_settings(
platform, app_name, 'align')
src_align_frames_path = src_align_frames_dict.get('align_src')
fill_format = '%05d.'
frames_file_pattern = os.path.join(image_basedir,
video_name + fill_format + frame_type)
src_frames_file_pattern = os.path.join(
# src_align_frames_path, 'default' + fill_format + frame_type
src_align_frames_path,
'0' + fill_format + frame_type)
x = roi_setting.get('x')
y = roi_setting.get('y')
width = roi_setting.get('width')
height = roi_setting.get('height')
if pre_roi_frames_dict:
pre_x = pre_roi_frames_dict.get('x')
pre_y = pre_roi_frames_dict.get('y')
pre_width = pre_roi_frames_dict.get('width')
pre_height = pre_roi_frames_dict.get('height')
resize_width = resize_frames_dict.get('width')
resize_height = resize_frames_dict.get('height')
if not width:
width = video_shape[0] - x
if not height:
height = video_shape[1] - y
logger.info('use frame rate:{0} to extract video'.format(frame_rate))
frame_generator = VideoFrameGenerator(video_file_path, video_shape,
frame_rate)
with frame_generator as frames:
i = 1
prev_frame = None
# 数字区域坐标信息
block_info = []
with open(error_csv_output_path,
'wb') as error_csv, open(fr_csv_output_path, 'wb') as fr_csv:
error_csv_writer = csv.writer(error_csv)
fr_csv_writer = csv.writer(fr_csv)
for frame in frames:
should_write = True
# 切黑边
if pre_roi_frames_dict:
pre_roi_filter = ROIFrameFilter(pre_x, pre_y, pre_width,
pre_height)
frame = pre_roi_filter.filter(frame)
# 放大
if video_shape[0] != resize_width or video_shape[
1] != resize_height:
resize_filter = ResizeFrameFilter(resize_width,
resize_height)
frame = resize_filter.filter(frame)
# 切广告和水印
roi_filter = ROIFrameFilter(x, y, width, height)
roi_frame = roi_filter.filter(frame)
if prev_frame is not None:
should_write = not numpy.array_equal(prev_frame, roi_frame)
if should_write:
frame_file_name = frames_file_pattern % (i, )
black_detector = AverageGrayThresholdDetector(0)
if int(black_detector.get_avg_gray(roi_frame)) > 20:
cv2.imwrite(frames_file_pattern % (i, ), roi_frame)
try:
# 识别失败会返回空字符串，int 强制转换会 raise exception
num = int(
detector.Detector(img=frame,
logger=logger).num_str)
src_path = src_frames_file_pattern % (num, )
if not os.path.exists(src_path):
logger.error('src pic %s is not exists.' %
src_path)
target_path = frame_file_name
src = cv2.imread(src_path, cv2.IMREAD_GRAYSCALE)
target = cv2.imread(target_path,
cv2.IMREAD_GRAYSCALE)
score = ssim(src, target, downsample=False)
row = [
target_path, src_path,
round(float(score), 5)
]
fr_csv_writer.writerow(row)
except Exception:
logger.exception('%s detect number failed' %
frame_file_name)
error_csv_writer.writerow([frame_file_name])
prev_frame = roi_frame
i += 1
return [
os.path.join(image_basedir, _) for _ in os.listdir(image_basedir)
if os.path.isfile(os.path.join(image_basedir, _))
]
def process_image_handler(error_csv_output_path, fr_csv_output_path,
video_file_path, image_basedir, roi_setting,
settings):
frame_type = 'bmp'
video_name = os.path.basename(video_file_path).split('.')[0]
platform = video_name.split('_')[1]
app_name = video_name.split('_')[0]
# TODO 把参数获取方式整理好，现在好恶心
pre_roi_frames_dict = None
try:
pre_roi_frames_dict = settings.get_specific_settings(
platform, app_name, 'pre_roi')
except ValueError as e:
logger.info('%s_%s no need pre roi' % (app_name, platform))
resize_frames_dict = settings.get_specific_settings(
platform, app_name, 'resize')
video_shape = get_video_shape(video_file_path)
# frame_rate = get_frame_rate(video_file_path)
frame_rate = 60
if not os.path.exists(image_basedir):
os.makedirs(image_basedir)
src_align_frames_dict = settings.get_specific_settings(
platform, app_name, 'align')
src_align_frames_path = src_align_frames_dict.get('align_src')
fill_format = '%05d.'
frames_file_pattern = os.path.join(image_basedir,
video_name + fill_format + frame_type)
src_frames_file_pattern = os.path.join(
# src_align_frames_path, 'default' + fill_format + frame_type
src_align_frames_path,
'0' + fill_format + frame_type)
x = roi_setting.get('x')
y = roi_setting.get('y')
width = roi_setting.get('width')
height = roi_setting.get('height')
if pre_roi_frames_dict:
pre_x = pre_roi_frames_dict.get('x')
pre_y = pre_roi_frames_dict.get('y')
pre_width = pre_roi_frames_dict.get('width')
pre_height = pre_roi_frames_dict.get('height')
resize_width = resize_frames_dict.get('width')
resize_height = resize_frames_dict.get('height')
if not width:
width = video_shape[0] - x
if not height:
height = video_shape[1] - y
logger.info('use frame rate:{0} to extract video'.format(frame_rate))
frame_generator = VideoFrameGenerator(video_file_path, video_shape,
frame_rate)
with frame_generator as frames:
i = 1
prev_frame = None
# 数字区域坐标信息
block_info = []
from contextlib import nested
with nested(open(error_csv_output_path, 'wb'),
open(fr_csv_output_path, 'wb')) as (error_csv, fr_csv):
error_csv_writer = csv.writer(error_csv)
fr_csv_writer = csv.writer(fr_csv)
for frame in frames:
should_write = True
# 切黑边
if pre_roi_frames_dict:
pre_roi_filter = ROIFrameFilter(pre_x, pre_y, pre_width,
pre_height)
frame = pre_roi_filter.filter(frame)
# 放大
if video_shape[0] != resize_width or video_shape[
1] != resize_height:
resize_filter = ResizeFrameFilter(resize_width,
resize_height)
frame = resize_filter.filter(frame)
# 切广告和水印
roi_filter = ROIFrameFilter(x, y, width, height)
roi_frame = roi_filter.filter(frame)
if prev_frame is not None:
should_write = not numpy.array_equal(prev_frame, roi_frame)
if should_write:
frame_file_name = frames_file_pattern % (i, )
# 数字识别，使用放大后的图（未切广告和水印）
if not block_info:
block_info = num_detector.detect_num_block(frame)
if block_info:
logger.info('detect block_info by %s success: %s' %
(frame_file_name, block_info))
else:
logger.error('detect block_info by %s failed.' %
frame_file_name)
black_detector = AverageGrayThresholdDetector(0)
if int(black_detector.get_avg_gray(roi_frame)) > 2:
cv2.imwrite(frames_file_pattern % (i, ), roi_frame)
num_str = num_detector.detect(frame, block_info)
# print num_str
try:
num = int(num_str)
# 黑帧不写入对帧结果，有效帧范围：301~7506
if 300 < num < 7507:
src_path = src_frames_file_pattern % (num, )
target_path = frame_file_name
src = cv2.imread(src_path,
cv2.IMREAD_GRAYSCALE)
target = cv2.imread(target_path,
cv2.IMREAD_GRAYSCALE)
score = ssim(src, target, downsample=False)
row = [
target_path, src_path,
round(float(score), 5)
]
# detect_csv_writer.writerow(row)
fr_csv_writer.writerow(row)
except ValueError as e:
logger.error('%s detect number failed' %
frame_file_name)
error_csv_writer.writerow([frame_file_name])
prev_frame = roi_frame
i += 1
return [
os.path.join(image_basedir, _) for _ in os.listdir(image_basedir)
if os.path.isfile(os.path.join(image_basedir, _))
]
def main():
parser = argparse.ArgumentParser()
parser.add_argument('-v',
'--verbose',
action='store_true',
default=False,
help='verbose')
parser.add_argument('-r',
'--frame-rate',
type=float,
help='frame rate',
default='1.0')
parser.add_argument('-d',
'--distinct',
action='store_true',
default=False,
help='distinct')
parser.add_argument('-t', '--frame-type', help='frame type')
parser.add_argument('video_dir', help='video\'s dir')
parser.add_argument('output_disk', help='output disk')
parser.add_argument('business_type',
help="pron:0, ad_orc:1, game_ocr:2",
choices=['0', '1', '2'])
parser.add_argument('col_int_01',
help='game type or something else',
choices=[
'juediqiusheng', 'huangyexingdong',
'cijizhanchang', 'quanjunchuji', 'wangzherongyao'
])
args = parser.parse_args()
frame_rate = args.frame_rate
distinct = args.distinct
frame_type = args.frame_type if args.frame_type is not None else 'jpg'
output_disk = args.output_disk
business_type = args.business_type
col_int_01 = args.col_int_01
if not os.path.exists(args.video_dir):
raise RuntimeError('video dir ' + args.video_dir + 'not exists')
if not os.path.exists(output_disk):
raise RuntimeError('output disk ' + args.output_disk + 'not exists')
# prepare video
video_metas = []
video_metas_queue = Queue.Queue()
for root, dirs, files in os.walk(args.video_dir):
if files:
for file_ in files:
# print file_
if file_.lower().endswith(deal_ext):
video_metas_queue.put(
VideoMeta(os.path.join(root, file_), output_disk))
video_meta = None
try:
while video_metas_queue.qsize() > 0:
video_meta = video_metas_queue.get()
print("undeal video %s" % (video_metas_queue.qsize()))
if not os.path.exists(video_meta.frames_file_dir):
os.makedirs(video_meta.frames_file_dir)
if os.path.exists(video_meta.meta_path) and os.path.isfile(
video_meta.meta_path):
print('video %s had deal' % (video_meta.video_path, ))
continue
video_shape = get_video_shape(video_meta.video_path)
frames_dirs = extract_frame(video_meta,
video_shape,
video_meta.frames_file_pattern,
frame_rate=frame_rate,
distinct=distinct)
with open(video_meta.meta_path, 'wb') as meta_file:
meta_file.writelines(frames_dirs)
save_to_database(video_meta.meta_path, business_type,
GameType[col_int_01].value)
except Exception:
if os.path.exists(video_meta.meta_path):
os.remove(video_meta.meta_path)
print traceback.format_exc()