基于深度学习的车牌检测系统的设计与实现（安卓、YOLOV、CRNNLPRNet）+文档

类型

图片数

说明

ccpd_base

199998

正常车牌

ccpd_challenge

10006

比较有挑战的车牌

ccpd_db

20001

光线较暗或较亮车牌

ccpd_fn

19999

距离摄像头较远或较近

ccpd_np

3036

没上牌的新车

ccpd_rotate

9998

水平倾斜20-50度，垂直倾斜-10-10度

ccpd_tilt

10000

水平倾斜15-45度，垂直倾斜-15-45度

ccpd_weather

9999

雨天、雪天或大雾的车牌

命名

具体含义

025

车牌区域占整个画面的比例

95_113

车牌水平和垂直角度, 水平95°, 竖直113°

154&383_386&473

标注框左上、右下坐标，左上(154, 383), 右下(386, 473)

86&473_177&454_154&383_363&402

标注框四个角点坐标，顺序为右下、左下、左上、右上

025

车牌区域占整个画面的比例；

0_0_22_27_27_33_16

第一个0为省份 对应省份字典provinces中的’皖’,；第二个0是该车所在地的地市一级代码，对应地市一级代码字典alphabets的’A’；后5位为字母和文字, 查看车牌号ads字典，如22为Y，27为3，33为9，16为S，最终车牌号码为皖AY339S。

def get_plate_licenses(plate):

    provinces = ["皖", "沪", "津", "渝", "冀", "晋", "蒙", "辽", "吉", "黑", "苏", "浙", "京", "闽", "赣", "鲁", "豫", "鄂", "湘", "粤",

                 "桂", "琼", "川", "贵", "云", "藏", "陕", "甘", "青", "宁", "新", "警", "学", "O"]

    alphabets = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'J', 'K', 'L', 'M', 'N', 'P', 'Q', 'R', 'S', 'T', 'U', 'V',

                 'W', 'X', 'Y', 'Z', 'O']

    ads = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'J', 'K', 'L', 'M', 'N', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X',

           'Y', 'Z', '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', 'O']

    result = [provinces[int(plate[0])], alphabets[int(plate[1])]]

    result += [ads[int(p)] for p in plate[2:]]

    result = "".join(result)

    # 新能源车牌的要求，如果不是新能源车牌可以删掉这个if

    # if result[2] != 'D' and result[2] != 'F' \

    #         and result[-1] != 'D' and result[-1] != 'F':

    #     print(plate)

    #     print("Error label, Please check!")

    print(plate, result)

    return result

def 比较图片大小(dir_image1, dir_image2):

    with open(dir_image1, "rb") as f1:

        size1 = len(f1.read())

    with open(dir_image2, "rb") as f2:

        size2 = len(f2.read())

    if (size1 == size2):

        result = "大小相同"

    else:

        result = "大小不同"

    return result

def 比较图片尺寸(dir_image1, dir_image2):

    image1 = Image.open(dir_image1)

    image2 = Image.open(dir_image2)

    if (image1.size == image2.size):

        result = "尺寸相同"

    else:

        result = "尺寸不同"

    return result

def 比较图片内容(dir_image1, dir_image2):

    image1 = np.array(Image.open(dir_image1))

    image2 = np.array(Image.open(dir_image2))

    if (np.array_equal(image1, image2)):

        result = "内容相同"

    else:

        result = "内容不同"

    return result

def 比较两张图片是否相同(dir_image1, dir_image2):

    result = "两张图不同"

    大小 = 比较图片大小(dir_image1, dir_image2)

    if (大小 == "大小相同"):

        尺寸 = 比较图片尺寸(dir_image1, dir_image2)

        if (尺寸 == "尺寸相同"):

            内容 = 比较图片内容(dir_image1, dir_image2)

            if (内容 == "内容相同"):

                result = "两张图相同"

    return result

def getImageVar(imgPath):

    image = cv2.imread(imgPath)

    img2gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

    imageVar = cv2.Laplacian(img2gray, cv2.CV_64F).var()

    return imageVar

if __name__ == "__main__":

    src_img_dir = r"D:\data\img_dir"

    move_img_dir = r"D:\data\img_dir_blur"

    os.makedirs(move_img_dir, exist_ok=True)

    img_files = os.listdir(src_img_dir)

    print("len(img_files): ", len(img_files))

    a_list = []

    for i in range(len(img_files)):

        img_file_path = os.path.join(src_img_dir, img_files[i])

        imageVar = getImageVar(img_file_path)

        # print(imageVar)

        a = item()

        a.val = imageVar

        a.name = img_files[i]

        a_list.append(a)

    print("len(a_list): ", len(a_list))

    a_list.sort(key=lambda ita: ita.val, reverse=False) # 按照 图片清晰度排序，模糊的放在列表头部，清晰的放在列表尾部

    count = 0

    for i in range(int(len(a_list)*0.1)): # 移除最模糊的 %10 的图片

        print(a_list[i].name, a_list[i].val)

        src_path = os.path.join(src_img_dir, a_list[i].name)

        dest_path = os.path.join(move_img_dir, a_list[i].name)

        shutil.move(src_path, dest_path)

        count += 1

        # break

    print("count: ", count)

def flip(root_path,img_name):   #翻转图像

    img = Image.open(os.path.join(root_path, img_name))

    filp_img = img.transpose(Image.FLIP_LEFT_RIGHT)

    # filp_img.save(os.path.join(root_path,img_name.split('.')[0] + '_flip.jpg'))

    return filp_img

def rotation(root_path, img_name):

    img = Image.open(os.path.join(root_path, img_name))

    random_angle = np.random.randint(-2, 2)*90

    if random_angle==0:

     rotation_img = img.rotate(-180) #旋转角度

    else:

        rotation_img = img.rotate( random_angle,expand=True)  # 旋转角度

    # rotation_img.save(os.path.join(root_path,img_name.split('.')[0] + '_rotation.jpg'))

    return rotation_img

def enhance_mycode(img):

    imgHeight,imgWidth,imgDeep = img.shape

    dst1 = np.zeros((imgHeight, imgWidth, 3), np.uint8)

    for i in range(0, imgHeight):

        for j in range(0, imgWidth):

            (b, g, r) = map(int,img[i,j])

            b += 40

            g += 40

            r += 40

            if b > 255:

                b = 255

            if g > 255:

                g = 255

            if r > 255:

                r = 255

            dst1[i, j] = (b, g, r)

    return dst1

def enhance_api(img):

    dst2 = np.uint8(np.clip((1.5 * img ), 0, 255))

    return dst2

from sklearn.model_selection import train_test_split

#使⽤train_test_split划分训练集和测试集

train_X , test_X, train_Y ,test_Y = train_test_split(

                 X, Y, test_size=0.2,random_state=0)

def ConverCoordinate(imgshape, bbox):

    # 将xml像素坐标转换为txt归一化后的坐标

    xmin, xmax, ymin, ymax = bbox

    width = imgshape[0]

    height = imgshape[1]

    dw = 1. / width

    dh = 1. / height

    x = (xmin + xmax) / 2.0

    y = (ymin + ymax) / 2.0

    w = xmax - xmin

    h = ymax - ymin

    # 归一化

    x = x * dw

    y = y * dh

    w = w * dw

    h = h * dh

    return x, y, w, h