AI视频数字识别、分割（上）
作者：互联网
任务描述:
根据视频中的数字的变化切割视频（国网需求）；
夜晚热红外视频帧：
根据右上角机位数字分割视频
思路:
- 视频分帧
- aoi区域裁剪
- 字符分割
- 神经网络模型训练
- 字符识别
- 断点确定
- 分割视频
一、视频分帧
# Step1:视频读取、分帧
# 1.保存视频每一秒的图面
def save_image(image, addr, num):
address = addr + str(num) + '.jpg'
cv2.imencode('.jpg', image)[1].tofile(address)
# cv2.imwrite(address, image)
# 2.视频分帧
def video_clip(src_path, save_path):
# 读取视频
video_capture = cv2.VideoCapture(src_path)
# 读帧
success, frame = video_capture.read()
# 设置固定帧率（查看视频属性）
timeF = 25
i = 0
j = 0
while success:
i = i + 1
if (i % timeF == 0):
j = j + 1
save_image(frame, save_path, j)
success, frame = video_capture.read()
print("================视频分帧完成！===================")
输出：
二、aoi区域裁剪
# Step2:AOI切割
# 1.选择裁剪区域
def select_point(frame_path):
frame_img = Image.open(frame_path + '1.jpg')
plt.imshow(frame_img, cmap=plt.get_cmap("gray"))
aoi_points = plt.ginput(4) # 选4个点（左上，右上，左下，右下）
return aoi_points
# 2.获取四至坐标
def get_bound(aoi_pts):
x_arr = []
y_arr = []
for pt in aoi_pts:
x_arr.append(int(pt[0]))
y_arr.append(int(pt[1]))
return min(x_arr), max(x_arr), min(y_arr), max(y_arr)
# 3.批量裁剪
def clip_frame_img(frame_path, aoi_path):
# 获取裁剪范围
aoi_pts = select_point(frame_path)
x_min, x_max, y_min, y_max = get_bound(aoi_pts)
# 裁剪
frame_name = os.listdir(frame_path)
print("正在进行aoi区域裁剪，请稍候...")
for frame in frame_name:
frame_full_path = os.path.join(frame_path, frame)
frame_image = cv2.imdecode(np.fromfile(frame_full_path, dtype=np.uint8), -1)
aoi_image = frame_image[y_min:y_max, x_min:x_max]
cv2.imencode('.jpg', aoi_image)[1].tofile(aoi_path + frame)
print("================aoi区域裁剪完成！===================")
# 没跳出裁剪窗口？
# pycharm设置问题，修改下配置：
# windows系统：Settings | Tools | Python Scientific | Show Plots in Toolwindow，去掉对勾
# MAC系统：Preferences | Tools | Python Scientific | Show Plots in Toolwindow，去掉对勾
输出：
三、字符分割
参照：https://www.cnblogs.com/zxy-joy/p/10687152.html
# Step3:字符分割
# 水平投影
def horizon_prj(image):
h_prj = np.zeros(image.shape, np.uint8)
# 图像高与宽
(h, w) = image.shape
# 长度与图像高度一致的数组
h_ = [0] * h
# 循环统计每一行黑色像素的个数（image[y, x] == 255是白色）
for y in range(h):
for x in range(w):
if image[y, x] == 0:
h_[y] += 1
# 绘制水平投影图像
for y in range(h):
for x in range(h_[y]):
h_prj[y, x] = 0
return h_
# 垂直投影
def vetical_project(image):
v_prj = np.zeros(image.shape, np.uint8) # 创建空白图片
(h, w) = image.shape
# 长度与图像宽度一致的数组
w_ = [0] * w
# 循环统计每一列白色像素的个数
for x in range(w):
for y in range(h):
if image[y, x] == 0:
w_[x] += 1
# 绘制垂直平投影图像
for x in range(w):
for y in range(h - w_[x], h):
v_prj[y, x] = 0
return w_
# aoi投影
def aoi_prj(aoi_image):
aoi_b2_image = cv2.cvtColor(aoi_image, cv2.COLOR_BGR2GRAY)
retval, img = cv2.threshold(aoi_b2_image, 127, 255, cv2.THRESH_BINARY_INV)
h, w = img.shape
position = []
# 水平投影
H = horizon_prj(img)
h_start = 0
H_Start = []
H_End = []
for i in range(len(H)):
if H[i] > 0 and h_start == 0:
H_Start.append(i)
h_start = 1
if H[i] <= 0 and h_start == 1:
H_End.append(i)
h_start = 0
if len(H_End) == 0:
H_End.append(h)
# 分割行，分割之后再进行列分割并保存分割位置
for i in range(len(H_Start)):
# 获取行图像
crop_img = img[H_Start[i]:H_End[i], 0:w]
# 对行图像进行垂直投影
W = vetical_project(crop_img)
w_start = 0
w_end = 0
W_Start = 0
W_End = 0
for j in range(len(W)):
if W[j] > 0 and w_start == 0:
W_Start = j
w_start = 1
w_end = 0
if W[j] <= 0 and w_start == 1:
W_End = j
w_start = 0
w_end = 1
if w_end == 1:
position.append([W_Start, H_Start[i], W_End, H_End[i]])
w_end = 0
# 根据确定的位置分割字符
# for m in range(len(position)):
# cv2.rectangle(aoi_image, (position[m][0], position[m][1]), (position[m][2], position[m][3]),
# (0, 229, 238),
# 1)
return position
# 字符分割（个人逻辑，仅供参考）
# 分割难免有误，或多或少
# 根据视频的连续性，多的去之，少的替之（以前后帧相替）
def character_split(aoi_path, character_path):
aoi_name = os.listdir(aoi_path)
all_aoi_positions = []
for aoi in aoi_name:
aoi_full_path = os.path.join(aoi_path, aoi)
aoi_image = cv2.imdecode(np.fromfile(aoi_full_path, dtype=np.uint8), -1)
aoi_positions = aoi_prj(aoi_image)
i = 0
if len(aoi_positions) == 7:
all_aoi_positions.append(aoi_positions)
elif len(aoi_positions) == 8:
del (aoi_positions[0])
else:
if i == 0:
print("第一个就有问题！")
else:
aoi_positions = all_aoi_positions[i - 1]
l = 0
for aoi_position in aoi_positions:
crop_image = aoi_image[aoi_position[1]:aoi_position[3], aoi_position[0]:aoi_position[2]]
crop_2image = np.where(crop_image[..., :] < 127, 0, 255)
row_crop_2image = crop_2image[1, :]
row_mean = row_crop_2image.mean(axis=1).tolist()
if row_mean.count(255.0) > len(row_mean) * 2 / 3:
crop_2image = np.where(crop_2image[..., :] == 0, 255, 0)
# print(row_mean.count(255.0))
# cv2.imencode('.jpg', crop_2image)[1].tofile(character_path + aoi.split('.')[0] + '_' + str(i) + '.jpg')
if l != 4:
cv2.imencode('.jpg', crop_2image)[1].tofile(
character_path + aoi.split('.')[0] + '_' + str(l) + '.jpg')
l += 1
i = i + 1
print("================字符分割完成！===================")
输出：
内容有点多，其他的放下一波吧~
标签：视频,分割,AI,image,cv2,aoi,path,frame 来源： https://blog.csdn.net/m0_37970224/article/details/114916540