Realization of Vehicle License Plate Recognition System (2)-License Plate Location
Preface
I am a rookie in computer vision processing in my senior year. Because of my personal interest and completion, I want to learn about computer vision. I am waiting for your criticisms and corrections for errors.
Image preprocessing
When the camera acquires an image of a car license plate, in most cases, it is unrealistic that there is only one license plate in the image. It is normal that there are buildings, trees, other cars, etc. in an image; and it is extremely large. Most of them will have noise points, in this case it will cause great interference to the license plate positioning and recognition!
The principles of Gaussian smoothing and median filtering have been introduced very clearly in many blogs on the Internet, so I won’t say more here, just go to the code!
** Note: Here is just a reminder, everyone must distinguish between black on white or white on black. I suffered a lot when I was studying. This is very important! **
#对图像进行缩放处理
if img_width > MAX_WIDTH:
resize_rate = MAX_WIDTH / img_width
img = cv2.resize(img,(MAX_WIDTH,int(img_hight * resize_rate)),interpolation=cv2.INTER_AREA)
# cv_show('img_resize',img)
# 高斯平滑
img_aussian = cv2.GaussianBlur(img,(5,5),1)
# cv_show('img_aussian',img_aussian)
#中值滤波
img_median = cv2.medianBlur(img_aussian,3)
# cv_show('img_median',img_median)
License plate characteristics
The object of our research is blue and white license plates, other license plates cannot be recognized temporarily. Having said that, by the way, explain the color information of the blue license plate and white license plate. For a normally exposed image, the three channel values of the blue license plate are approximately Blue=138, Green=63, Red=23. In addition to the color information, the license plate has a rectangular shape with a fixed aspect ratio of 3:1. (This paragraph is taken from https://blog.csdn.net/qq_39960119/article/details/83930112)
License plate positioning
License plate location is usually also called license plate extraction, and its purpose is to find the license plate area from the image containing the license plate. The importance of license plate location is self-evident. As the first step of license plate recognition, the success or failure of the location of the license plate area is the foundation of the license plate recognition system and the primary determining factor. This system uses a positioning method based on color information. This method has advantages and disadvantages. The advantage is that it can quickly and accurately locate the license plate area. The disadvantage is that the light intensity and weather have a great influence on the positioning effect. However, for novices, this method It is simpler and easier to use.
The code has just been completed, it's messy, it will be sorted and functionalized later. I try to comment as much as possible. It is convenient for everyone to read. If you don’t understand, you can comment and leave a message.
#分离通道
img_B = cv2.split(img_median)[0]
img_G = cv2.split(img_median)[1]
img_R = cv2.split(img_median)[2]
# print(img_B)
# print(img_G)
# print(img_R)
for i in range(img_median.shape[:2][0]):
for j in range(img_median.shape[:2][1]):
if abs(img_B[i,j] - Blue) < THRESHOLD and abs(img_G[i,j] - Green) <THRESHOLD and abs(img_R[i,j] - Red) < THRESHOLD:
img_median[i][j][0] = 255
img_median[i][j][1] = 255
img_median[i][j][2] = 255
else:
img_median[i][j][0] = 0
img_median[i][j][1] = 0
img_median[i][j][2] = 0
cv_show('img_median',img_median)
kernel = np.ones((3,3),np.uint8)
img_dilate = cv2.dilate(img_median,kernel,iterations = 5) #膨胀操作
img_erosion = cv2.erode(img_dilate,kernel,iterations = 5) #腐蚀操作
cv_show('img_erosion',img_erosion)
img1 = cv2.cvtColor(img_erosion,cv2.COLOR_RGB2GRAY)
# cv_show('img1',img1)
image, contours, hierarchy = cv2.findContours(img1,cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE)
# for i in range(len(contours)):
# print(contours[i])
car_contours = []
for cnt in contours: #TODO：此处需要一个异常处理（因为有可能/0）
# 框选 生成最小外接矩形 返回值（中心(x,y), (宽,高), 旋转角度）rect[0]：矩形中心点坐标；rect[1]：矩形的高和宽；rect[2]：矩形的旋转角度
rect = cv2.minAreaRect(cnt)
# print('宽高:',rect[1])
area_width, area_height = rect[1]
#计算最小矩形的面积，初步筛选
area = rect[1][0] * rect[1][1] #最小矩形的面积
if area > MIN_AREA:
# 选择宽大于高的区域
if area_width < area_height:
area_width, area_height = area_height, area_width
wh_ratio = area_width / area_height
# print('宽高比：',wh_ratio)
# 要求矩形区域长宽比在2到5.5之间，2到5.5是车牌的长宽比，其余的矩形排除
if wh_ratio > 2 and wh_ratio < 5.5:
car_contours.append(rect) # rect是minAreaRect的返回值，根据minAreaRect的返回值计算矩形的四个点
box = cv2.boxPoints(rect) # box里面放的是最小矩形的四个顶点坐标
box = np.int0(box) # 取整
# for i in range(len(box)):
# print('最小矩形的四个点坐标：', box[i])
# 获取四个顶点坐标
left_point_x = np.min(box[:, 0])
right_point_x = np.max(box[:, 0])
top_point_y = np.min(box[:, 1])
bottom_point_y = np.max(box[:, 1])
left_point_y = box[:, 1][np.where(box[:, 0] == left_point_x)][0]
right_point_y = box[:, 1][np.where(box[:, 0] == right_point_x)][0]
top_point_x = box[:, 0][np.where(box[:, 1] == top_point_y)][0]
bottom_point_x = box[:, 0][np.where(box[:, 1] == bottom_point_y)][0]
vertices = np.array([[top_point_x, top_point_y], [bottom_point_x, bottom_point_y], [left_point_x, left_point_y],[right_point_x, right_point_y]])
oldimg = cv2.drawContours(img, [box], 0, (0, 0, 255), 2)
# print(car_contours)
cv_show('oldimg', oldimg)
Preprocessing and license plate positioning all code
** If the environment is OK, it should be able to run after copying! **
import numpy as np
import cv2
import argparse
from matplotlib import pyplot as plt
Blue = 138
Green = 63
Red = 23
THRESHOLD = 50
ANGLE = -45
MIN_AREA = 2000
LICENSE_WIDTH = 440
LICENSE_HIGH = 140
#绘图展示(后期放入工具类中)
def cv_show(name,img):
cv2.namedWindow(name,0)
print(img.shape)
cv2.imshow(name,img)
cv2.waitKey(0)
cv2.destroyAllWindows()
MAX_WIDTH = 640
# 设置参数
parser = argparse.ArgumentParser()
parser.add_argument("-i", "--image", required=True,help="path to input image") #image是变量名
args = vars(parser.parse_args())
#读取原图像
img = cv2.imread(args["image"])
cv_show('img',img)
img_hight, img_width = img.shape[:2]
#-------------------------------预处理--------------------------------------
#对图像进行缩放处理
if img_width > MAX_WIDTH:
resize_rate = MAX_WIDTH / img_width
img = cv2.resize(img,(MAX_WIDTH,int(img_hight * resize_rate)),interpolation=cv2.INTER_AREA)
# cv_show('img_resize',img)
# 高斯平滑
img_aussian = cv2.GaussianBlur(img,(5,5),1)
# cv_show('img_aussian',img_aussian)
#中值滤波
img_median = cv2.medianBlur(img_aussian,3)
# cv_show('img_median',img_median)
# print('width：',img_median.shape[:2][0])
# print('h',img_median.shape[:2][1])
#------------------------------车牌定位-------------------------------------
#分离通道
img_B = cv2.split(img_median)[0]
img_G = cv2.split(img_median)[1]
img_R = cv2.split(img_median)[2]
# print(img_B)
# print(img_G)
# print(img_R)
for i in range(img_median.shape[:2][0]):
for j in range(img_median.shape[:2][1]):
if abs(img_B[i,j] - Blue) < THRESHOLD and abs(img_G[i,j] - Green) <THRESHOLD and abs(img_R[i,j] - Red) < THRESHOLD:
img_median[i][j][0] = 255
img_median[i][j][1] = 255
img_median[i][j][2] = 255
else:
img_median[i][j][0] = 0
img_median[i][j][1] = 0
img_median[i][j][2] = 0
cv_show('img_median',img_median)
kernel = np.ones((3,3),np.uint8)
img_dilate = cv2.dilate(img_median,kernel,iterations = 5) #膨胀操作
img_erosion = cv2.erode(img_dilate,kernel,iterations = 5) #腐蚀操作
cv_show('img_erosion',img_erosion)
img1 = cv2.cvtColor(img_erosion,cv2.COLOR_RGB2GRAY)
# cv_show('img1',img1)
image, contours, hierarchy = cv2.findContours(img1,cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE)
# for i in range(len(contours)):
# print(contours[i])
car_contours = []
for cnt in contours: #TODO：此处需要一个异常处理（因为有可能/0）
# 框选 生成最小外接矩形 返回值（中心(x,y), (宽,高), 旋转角度）rect[0]：矩形中心点坐标；rect[1]：矩形的高和宽；rect[2]：矩形的旋转角度
rect = cv2.minAreaRect(cnt)
# print('宽高:',rect[1])
area_width, area_height = rect[1]
#计算最小矩形的面积，初步筛选
area = rect[1][0] * rect[1][1] #最小矩形的面积
if area > MIN_AREA:
# 选择宽大于高的区域
if area_width < area_height:
area_width, area_height = area_height, area_width
wh_ratio = area_width / area_height
# print('宽高比：',wh_ratio)
# 要求矩形区域长宽比在2到5.5之间，2到5.5是车牌的长宽比，其余的矩形排除
if wh_ratio > 2 and wh_ratio < 5.5:
car_contours.append(rect) # rect是minAreaRect的返回值，根据minAreaRect的返回值计算矩形的四个点
box = cv2.boxPoints(rect) # box里面放的是最小矩形的四个顶点坐标
box = np.int0(box) # 取整
# for i in range(len(box)):
# print('最小矩形的四个点坐标：', box[i])
# 获取四个顶点坐标
left_point_x = np.min(box[:, 0])
right_point_x = np.max(box[:, 0])
top_point_y = np.min(box[:, 1])
bottom_point_y = np.max(box[:, 1])
left_point_y = box[:, 1][np.where(box[:, 0] == left_point_x)][0]
right_point_y = box[:, 1][np.where(box[:, 0] == right_point_x)][0]
top_point_x = box[:, 0][np.where(box[:, 1] == top_point_y)][0]
bottom_point_x = box[:, 0][np.where(box[:, 1] == bottom_point_y)][0]
vertices = np.array([[top_point_x, top_point_y], [bottom_point_x, bottom_point_y], [left_point_x, left_point_y],[right_point_x, right_point_y]])
oldimg = cv2.drawContours(img, [box], 0, (0, 0, 255), 2)
# print(car_contours)
cv_show('oldimg', oldimg)
operation result
Experimented with a few images, the effect is not bad!
1. Read the original image
2. After preprocessing and positioning method based on color information
3. After morphological operation
4. License plate positioning and filtering based on characteristics The
next article is about image correction!