CCPD data set processing (target detection and text recognition)

---

Preface

   lately , Yes CCPD Data sets , But open source CCPD Datasets are not easy to use , So I wrote a script with reference to the explanation of image format to generate a script that supports target detection and text recognition label file . All of the following have been CCPD2019 For example .

---

One 、CCPD Data set introduction

  CCPD The data set is mainly collected from the data of a parking lot in Anhui Province for a period of time , All picture sizes are fixed to 720×1160（w×h）, It contains about 25w+ Pictures of various scenes , As shown in the figure below ：

meanwhile , Also based on CCPD-Base Split the training set 、 Test set 、 Verification set , The file path is stored in splits Next , As shown in the figure ：

The remaining text corresponds to the relative path of the picture in each scene , It is convenient for us to divide , It is suggested that you divide the data set according to your own needs .

Two 、 Download address

   Download at ：CCPD

3、 ... and 、 Format interpretation

   After the image is unzipped, it looks like this ,

We can see that the file names here are all composed of a pile of numbers and symbols , What do these represent ？ Let's take the first file name ：“01-86_91-298&341_449&414-458&394_308&410_304&357_454&341-0_0_14_28_24_26_29-124-24.jpg” For example , according to “-” Separate these symbols from numbers , Each paragraph represents a meaning ：
 1、01: The proportion of license plates in the whole interface ;（ It doesn't work , Negligible ）
 2、86_91： The horizontal and vertical angles of the license plate
 3、298&341_449&414： Coordinates of the upper left corner and lower right corner of the license plate marking box
 4、458&394_308&410_304&357_454&341： The coordinates of the four vertices of the license plate , The order is bottom right 、 The lower left 、 Top left 、 The upper right
 5、0_0_14_28_24_26_29： This represents and provinces （ first place ）、 Local market （ Second ）、 license plate number （ rest ） The mapping relation of
 6、124： brightness , The higher the value, the higher the brightness （ For reference only ）
 7、24： Ambiguity , The smaller the value, the more blurred （ For reference only ）

Four 、 Generate corresponding lable

   Let's go directly to the code , You need to specify your own path in the code , Such as image file path and generation txt The path to save ; secondly , You also need to turn on the corresponding switch according to your task , If the license plate is located use_landmarks Set to True, Other similar ：

import os
import glob
import cv2
from pathlib import Path
import numpy as np

root_dir=""
txt_dir=""
use_landmarks=False # For license plate location , Such as MTCNN、YOLO5Face
use_xywh=False # For license plate detection , Such as YOLO etc. 
use_ocr=False # Used to identify 
os.makedirs(txt_dir,exist_ok=True)

# Returns the coordinates of the center point 、 wide 、 The height and the position information of the four vertices are 12 Messages , Can be used for mtcnn or yolo5Face Detect and locate the license plate 
def ccpd_landmarks(points_list,imgw,imgh):
    ''' Args: points_list:  The position information of the four vertices of the license plate  imgw:  Image width  imgh:  Picture height  Returns: '''
    annotation=np.zeros((1,12))
    points_list_=[]
    # Get four endpoints 
    for points in points_list.split("_"):
        points_=list(map(int,points.split("&")))
        points_list_.append(points_)
    points_list_=np.array(points_list_)

    # Acquired box Coordinates of 
    xmin=min(points_list_[:,0])
    xmax=max(points_list_[:,0])
    ymin=min(points_list_[:,1])
    ymax=max(points_list_[:,1])
    dw=1./imgw
    dh=1./imgh
    w,h=xmax-xmin,ymax-ymin
    # Generate the corresponding information list 
    annotation[0,0]=((xmin+xmax)/2.-1)*dw #cx
    annotation[0,1]=((ymin+ymax)/2.-1)*dh #cy
    annotation[0,2]=w*dw #w
    annotation[0,3]=h*dh #h
    annotation[0,4]=points_list_[2][0]/imgw # Top left x
    annotation[0,5]=points_list_[2][1]/imgh # Top left y
    annotation[0,6]=points_list_[3][0]/imgw # The upper right x
    annotation[0,7]=points_list_[3][1]/imgh # The upper right y
    annotation[0,8]=points_list_[0][0]/imgw # The lower right x
    annotation[0,9]=points_list_[0][1]/imgh # The upper right y
    annotation[0,10]=points_list_[1][0]/imgw # The lower left x
    annotation[0,11]=points_list_[1][1]/imgh # The lower left y
    return annotation


# Only the coordinates of the center point and the width and height information of the picture are returned , Can be used for YOLO Serial inspection license plate 
def ccpd_xywh(points_list,imgw,imgh):
    ''' Args: points_list:  The position information of the four vertices of the license plate  imgw:  Image width  imgh:  Picture height  Returns: '''
    annotation = np.zeros((1, 4))
    points_list_ = []
    #  Get four endpoints 
    for points in points_list.split("_"):
        points_ = list(map(int, points.split("&")))
        points_list_.append(points_)
    points_list_ = np.array(points_list_)

    # Acquired box Coordinates of 
    xmin=min(points_list_[:,0])
    xmax=max(points_list_[:,0])
    ymin=min(points_list_[:,1])
    ymax=max(points_list_[:,1])
    dw=1./imgw
    dh=1./imgh
    w,h=xmax-xmin,ymax-ymin
    #  Generate the corresponding information list 
    annotation[0, 0] = ((xmin + xmax) / 2. - 1) * dw  # cx
    annotation[0, 1] = ((ymin + ymax) / 2. - 1) * dh  # cy
    annotation[0, 2] = w * dw  # w
    annotation[0, 3] = h * dh  # h
    return annotation

# Return the corresponding text information , Can be used for ocr License plate recognition 
def ccpd_ocr(plate):
    ''' Args: plate:  The number in the file name  Returns: '''
    # License plate dictionary 
    # province 
    provinces = [" Wan ", " Shanghai ", " tianjin ", " chongqing ", " Ji ", " Jin ", " Mongolia ", " liao ", " ji ", " black ", " Sue ", " Zhejiang ", " Beijing ", " Fujian ", " Gan ", " Lu ", " Yu ", " E ", " hunan ", " guangdong ",
                 " guangxi ", " Joan ", " sichuan ", " your ", " cloud ", " hidden ", " shan ", " gump ", " green ", " ning ", " new ", " police ", " learn ", "O"]
    # Specific information 
    word_lists = ['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', '1', '2', '3', '4', '5', '6', '7', '8', '9','0']


    label_list=plate.split("_")
    print(len(label_list))
    result=""
    result+=provinces[int(label_list[0])]
    result+=word_lists[int(label_list[1])]
    result+=word_lists[int(label_list[2])]+word_lists[int(label_list[3])]+word_lists[int(label_list[4])]+\
            word_lists[int(label_list[5])]+word_lists[int(label_list[6])]
    return result


def ccpd2label(all_list):
    for imgpath in all_list:
        img=cv2.imread(imgpath)
        h,w=img.shape[:2]

        imgname=Path(imgpath).name
        points_list,plate_list=imgname.split("-")[3],imgname.split("-")[4]# Coordinates of the four key points [ The lower right 、 The lower left 、 Top left 、 The upper right ], License plate number mapping table 
        if use_landmarks:
            annotation=ccpd_landmarks(points_list,w,h) # Get to write txt Information about 
            txtname = imgname.replace(".jpg", ".txt")
            txt_path = os.path.join(txt_dir, txtname)
            str_label = "0"  #  The category defaults to 0
            with open(txt_path, "a+") as fw:
                for i in range(len(annotation[0])):
                    str_label = str_label + " " + str(annotation[0][i])
                fw.write(str_label)
        elif use_xywh:
            annotation=ccpd_xywh(points_list,w,h) # Get to write txt Information about 
            txtname = imgname.replace(".jpg", ".txt")
            txt_path = os.path.join(txt_dir, txtname)
            str_label = "0"  #  The category defaults to 0
            with open(txt_path, "a+") as fw:
                for i in range(len(annotation[0])):
                    str_label = str_label + " " + str(annotation[0][i])
                fw.write(str_label)
        elif use_ocr:
            ocr_label=ccpd_ocr(plate_list)
            txtname=imgname.replace(".jpg",".txt")
            txt_path=os.path.join(txt_dir,txtname)
            with open(txt_path,"a+") as fw:
                fw.write(ocr_label)

if __name__ == '__main__':
    image_list=glob.glob(root_dir+os.sep+"*.jpg")
    ccpd2label(image_list)

Reference blog :https://blog.csdn.net/luohenyj/article/details/117752120

---

summary

The above is the whole content of this article , If you have any questions , Welcome to comment or join QQ Group :995760755 communication .