Skip to main content

Yolo App - YOLOv5 Data Preparation

Shenzhen, China

  1. Prepare your Images and get Data
  2. Train your Tensorflow Model
  3. Use your Model to do Predictions
  4. Use Tesseract to Read Number Plates
  5. Flask Web Application
  6. Yolo v5 - Data Prep

I now have a clean dataset, a working model and a web application for testing. But the detection process is relatively slow and not suitable for real-time video detection. This is where YOLOv5 comes in.

Data Preparation​

There is only one problem with the data that was used to train the Tensorflow model. There I needed to define the bounding box around detect license plates byt the variables xmin, xmax, ymin, ymax. But YOLO expects an X & Y value for the center point of the region of interest and it's height & width.

Load Labels​

Let's start by importing the labels files for our images:

import numpy as np
import pandas as pd
from glob import glob
import xml.etree.ElementTree as xet
import cv2
import os
import shutil import copy

importdf = pd.read_csv('../labels.csv')
df.head()
    filepath 	                xmin 	xmax 	ymin 	ymax
0 	../resources/cars_170.xml 	224 	439 	77 	  167
1 	../resources/cars_171.xml 	416 	620 	536 	600
2 	../resources/cars_172.xml 	184 	325 	114 	148
3 	../resources/cars_173.xml 	154 	373 	91 	  149
4 	../resources/cars_174.xml 	131 	279 	213 	256

Parse XML Data​

I now need to convert these bbox values to center_x, center_y, width and height and normalize them to their image size. I can start by extracting the image width and height from the generated XML label:

<annotation>
	<folder>resources</folder>
	<filename>cars_1.jpg</filename>
	<path>/opt/yolo-app/resources/cars_1.jpg</path>
	<source>
		<database>Unknown</database>
	</source>
	<size>
		<width>1600</width>
		<height>1153</height>
		<depth>3</depth>
	</size>
	<segmented>0</segmented>
	<object>
		<name>number_plate</name>
		<pose>Unspecified</pose>
		<truncated>0</truncated>
		<difficult>0</difficult>
		<bndbox>
			<xmin>1085</xmin>
			<ymin>561</ymin>
			<xmax>1354</xmax>
			<ymax>683</ymax>
		</bndbox>
	</object>
</annotation>

The function to parse the image label is:

# Parsing XML labels
def xmlparsing(path):
    parser = xet.parse(path).getroot()
    image_path = '../' + parser.find('folder').text + '/' + parser.find('filename').text
    image_size = parser.find('size')
    width = int(image_size.find('width').text)
    height = int(image_size.find('height').text)
    
    return image_path, width, height

Now I can append the image width and height to my Pandas dataframe with:

# Take filepath from df and function to append
# image_path, width and height from XML label
df[['image_path','width','height']] = df['filepath'].apply(xmlparsing).apply(pd.Series)
df.head()
    filepath 	                  xmin 	xmax 	ymin 	ymax 	image_path 	                width height
0 	../resources/cars_170.xml 	224 	439 	77 	  167 	../resources/cars_170.jpeg 	500 	234
1 	../resources/cars_171.xml 	416 	620 	536 	600 	../resources/cars_171.jpeg 	1070 	907
2 	../resources/cars_172.xml 	184 	325 	114 	148 	../resources/cars_172.jpeg 	500 	333
3 	../resources/cars_173.xml 	154 	373 	91 	  149 	../resources/cars_173.jpeg 	500 	250
4 	../resources/cars_174.xml 	131 	279 	213 	256 	../resources/cars_174.jpeg 	414 	432

Calculate Bounding Box​

And now to getting the variables that are needed by Yolo:

# Calculate center_x, center_y, width and height of bounding box
# and normalize them to image size
df['center_x'] = (df['xmax'] + df['xmin'])/(2*df['width'])
df['center_y'] = (df['ymax'] + df['ymin'])/(2*df['height'])

df['bb_width'] = (df['xmax'] - df['xmin'])/df['width']
df['bb_height'] = (df['ymax'] - df['ymin'])/df['height']

df.head()
 	filepath         image_path 	      width height 	center_x 	center_y 	bb_width 	bb_height
0 .../cars_170.xml .../cars_170.jpeg 	500 	234 	  0.663000 	0.521368 	0.430000 	0.384615
1 .../cars_171.xml .../cars_171.jpeg 	1070 	907 	  0.484112 	0.626240 	0.190654 	0.070562
2 .../cars_172.xml .../cars_172.jpeg 	500 	333 	  0.509000 	0.393393 	0.282000 	0.102102
3 .../cars_173.xml .../cars_173.jpeg 	500 	250 	  0.527000 	0.480000 	0.438000 	0.232000
4 .../cars_174.xml .../cars_174.jpeg 	414 	432 	  0.495169 	0.542824 	0.357488 	0.099537

Split Testing and Training Data​

Divide image into files used for training and for testing:

# Take first 220 images for training
df_train = df.iloc[:220]
# Take remaining images for testing
df_test = df.iloc[220:]

Create labels for training images and copy everything into the trainings folder:

# Training Data
train_folder = '../data/train'

train_values = df_train[['image_path', 'center_x', 'center_y', 'bb_width', 'bb_height']].values

# Create label and copy images to folder
for fname, x, y, w, h in train_values:
    # Get filename from filepath
    image_name = os.path.split(fname)[-1]
    # Remove file extension
    label_name = os.path.splitext(image_name)[0]
    
    # Copy training images to train folder
    dst_image_path = os.path.join(train_folder, image_name)
    copy(fname,dst_image_path)
    
    # Create image label file
    label_values = f'0 {x} {y} {w} {h}' 
    label_path = os.path.join(train_folder, label_name + '.txt')
    with open(label_path , mode='w') as f:
        f.write(label_values)
        f.close()

And repeat this step for the testing images:

# Testing Data
test_folder = '../data/test'

test_values = df_test[['image_path', 'center_x', 'center_y', 'bb_width', 'bb_height']].values

# Create label and copy images to folder
for fname, x, y, w, h in test_values:
    # Get filename from filepath
    image_name = os.path.split(fname)[-1]
    # Remove file extension
    label_name = os.path.splitext(image_name)[0]
    
    # Copy training images to train folder
    dst_image_path = os.path.join(test_folder, image_name)
    copy(fname,dst_image_path)
    
    # Create image label file
    label_values = f'0 {x} {y} {w} {h}' 
    label_path = os.path.join(test_folder, label_name + '.txt')
    with open(label_path , mode='w') as f:
        f.write(label_values)
        f.close()

Plate Detection Flask App