opencv数据集制作

原始视频文件，需要进行标签和图象处理，才能得到效果较好的数据集。

1. 视频初处理

首先需要对原始视频进行截断，修改尺寸，并保存为易于读取的格式。

#create dataset
video_file = r'./test.mp4'
cap = cv2.VideoCapture(video_file)
fps = 24
num_frames = 4*60*fps
video_size = (1920,1080)
video_sub_file = r'./test_sub.avi'
fourcc = cv2.VideoWriter_fourcc(*"mp42")
video_writer = cv2.VideoWriter(video_sub_file, fourcc, fps, video_size)
for i in tqdm(range(num_frames)):
    res, frame = cap.read()
    if not res:
        print("camera error")
        break
    video_writer.write(frame)
cap.release()
print("load finish")

fps = 24
num_frames = 4*60*fps
video_sub_file = r'./test_sub.avi'
cap = cv2.VideoCapture(video_sub_file)
video_width = cap.get(cv2.CAP_PROP_FRAME_WIDTH)
video_heigth = cap.get(cv2.CAP_PROP_FRAME_HEIGHT)
print(video_width,video_heigth)
# print(cap.isOpened())
# print(cap.read())
save_dir = './dataset'
if not os.path.exists(save_dir):
    os.makedirs(save_dir)

for i in tqdm(range(num_frames)):
    res, image = cap.read()
    #print(res)
    minute = i//(60*fps)
    second = (i - minute*60*fps)//fps
    if not res:
        print("camera error")
        break
    mid_dir = os.path.join(save_dir, 'v_{0}_c_{1:0>2d}_{2:0>2d}_t'.format(minute+1, minute, second))
    if not os.path.exists(mid_dir):
        os.makedirs(mid_dir)
    frame_name = os.path.join(mid_dir, 'frame{:0>6d}.jpg'.format(i%fps+1))
    new_image = cv2.resize(image,(255,255))
    video_file = cv2.imwrite(frame_name, new_image)
cap.release()
print("dataset finish")

# img_path = './dataset/v_1_c_00_00_t/frame000001.jpg'
# img = cv2.imread(img_path)
# print(img.shape)

最后结果将原始视频调整为4min、24fps、（255，255）尺寸的的数据集。

dataset

2. 图象处理

3. 数据集读取（关键帧）