Detecting Potholes on Roads Using Python and YOLOv8

Potholes are one of the major issues on roads that can cause accidents and damage vehicles. Identifying and fixing them early can improve road safety significantly. In this article, we’ll explore how to create a Pothole Detection Project using Python and YOLOv8, a powerful object detection model. This project can detect potholes in both images and videos, providing a practical solution to identify these dangerous road defects efficiently.

In this guide, we’ll walk through the steps to detect potholes in road images and videos using deep learning and the YOLOv8 object detection model. Whether you want to use this for analyzing images or for real-time detection, this project will help you get started.

Let’s dive into the details!

Requirements and Installations

Before we start coding, let’s ensure Python (3.6 or later) is installed on your computer. If you don’t have Python, you can download it for free from https://www.python.org/downloads/.

Now download all the dependencies we require using the following commands:

pip install gitpython>=3.1.30
pip install matplotlib>=3.3
pip install numpy>=1.23.5
pip install opencv-python>=4.1.1
pip install pillow>=10.3.0
pip install psutil 
pip install PyYAML>=5.3.1
pip install requests>=2.32.0
pip install scipy>=1.4.1
pip install thop>=0.1.1
pip install torch>=1.8.0
pip install torchvision>=0.9.0
pip install tqdm>=4.64.0
pip install ultralytics>=8.2.34
pip install pandas>=1.1.4
pip install seaborn>=0.11.0
pip install setuptools>=65.5.1
pip install filterpy
pip install scikit-image
pip install lap

Alternative Installation

Installing the above utilities one by one might be a boring task. Instead, you can download the ‘requirements.txt‘ file containing all the dependencies above. Simply run the following command. It will automate the whole task in one go.

pip install -r requirements.txt

Training of YOLO Model on Custom Dataset

At the very first, we have to train our YOLO model. Please follow the steps below:

Download the Dataset

Download the pothole detection dataset from roboflow.com.

Now unzip the downloaded dataset. The folder should look like the following:

Training YOLOv8 Model with Custom Dataset using Colab

Open Google Colab, sign in with your Gmail account, and open a new notebook.

Now go to the ‘Runtime‘ menu, select ‘Change runtime type‘, choose ‘T4 GPU‘ for the Hardware accelerator, and save it.

Let’s check whether the GPU is running perfectly or not using the following command:

!nvidia-smi

The output should look like the following:

Next, install ultralytics on your colab workspace using the following command:

!pip install ultralytics

Now open your Google Drive and navigate to ‘My Drive.’ Now create a folder named ‘Datasets‘ under ‘My Drive’ and inside the ‘Datasets’ folder create one more folder ‘Pothole.’

Let’s open the unzipped dataset folder, select all items present there, and drop them into the ‘Pothole’ folder on Google Drive. It may take a while so wait until it is finished. The final ‘Pothole’ folder will look like the following:

Now open the ‘data.yaml‘ file in the text editor and modify the path variable to: “../drive/MyDrive/Datasets/Pothole” The final ‘data.yaml‘ file will look like the following:

Now, let’s go back to our Google Colab dashboard. You need to mount your Google Drive with the Colab. Insert the following command in a new cell and run it:

from google.colab import drive
drive.mount('/content/drive')

You should get a success message like this: “Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount(“/content/drive”, force_remount=True).”

Now we will start training our YOLO model with our pothole detection dataset. Again, create a new cell, insert the command below, and run it.

!yolo task=detect mode=train model=yolov8l.pt data=../content/drive/MyDrive/Datasets/Pothole/data.yaml epochs=100 imgsz=640

Here, ‘epochs=100‘ specifies the number of training epochs. An epoch is one complete pass through the entire training dataset. Here, the model will be trained for 100 epochs.

‘imgsz=640‘ sets the size of the input images that the model will be trained on. In this case, images will be resized to 640×640 pixels before being fed into the model.

The whole training can take around 1 – 2 hours even more to complete.

After, the completion of the training go to the ‘Files‘ section in your Colab dashboard and navigate through these folders: ‘runs’ -> ‘detect’ -> ‘train’ -> ‘weights’. Inside the ‘weights‘ folder you will see ‘best.pt‘ and ‘last.pt‘ these two files. Download ‘best.pt‘ from there.

Setting Up the Environment

Create a separate folder named “Pothole_Detector” for this project. Under this folder create two more folders named ‘Weights‘ and ‘Media‘ to store pre-trained YOLO models and images respectively.

Place the Downloaded YOLO Model

In the previous section, we trained our YOLO model with a custom pothole detection dataset and downloaded a file named ‘best.pt.’ Now place this file inside the ‘Weights’ folder.

Media Files

I have collected some suitable images from the internet for this Pothole Detection Project. You can get those using the ‘Download‘ button below. All you have to do is, download the zip file, unzip it, and place those images inside the ‘Media‘ folder.

Create Your Python Script

We’re almost at the end of setting up the environment. Now choose your favorite text editor and open the entire project folder ‘Pothole_Detector.’ Inside this folder, create a Python program file named ‘pothole_detector.py.‘ This is where you’ll write the code.

Your final project file hierarchy should look like the following:

Pothole_Detector/
├── Weights/
│   └── best.pt
├── Media/
│   └── pothole_1.jpeg
│   └── pothole_2.jpeg
│   └── pothole_3.jpg
│   └── pothole_4.jpeg
│   └── Potholes.mp4
├── pothole_detector.py
├── pothole_detector_video.py

The Program: Detecting Potholes in Images

Let’s start writing your code step-by-step and try to understand the logic.

Import Libraries

First, we need to import the necessary libraries. Here, ‘OpenCV‘ is used for image processing, ‘cvzone‘ helps draw bounding boxes, and ‘YOLO’ from the ‘ultralytics‘ library is used for object detection.

import cv2
import math
import cvzone
from ultralytics import YOLO

Load YOLO Model and Define Class Names

Next, load the YOLO model with the custom-trained weights and define the class names that YOLO can detect. Make sure you have downloaded the ‘best.pt‘ weights and placed them in the correct directory.

# Load YOLO model with custom weights
yolo_model = YOLO("Weights/best.pt")

# Define class names
class_labels = ['Pothole']

Load the Image

Now, load the image you want to process using OpenCV’s ‘imread‘ method.

# Load the image
image_path = "Media/pothole_1.jpeg"
img = cv2.imread(image_path)

Perform Object Detection

The YOLO model is used to detect objects in the loaded image.

# Perform object detection
results = yolo_model(img)

Draw Bounding Boxes and Labels

Now we will loop through the detected objects and draw bounding boxes around them. The confidence score and class label will also be displayed.

# Loop through the detections and draw bounding boxes
for r in results:
    boxes = r.boxes
    for box in boxes:
        x1, y1, x2, y2 = box.xyxy[0]
        x1, y1, x2, y2 = int(x1), int(y1), int(x2), int(y2)

        w, h = x2 - x1, y2 - y1
        
        conf = math.ceil((box.conf[0] * 100)) / 100
        cls = int(box.cls[0])

        if conf > 0.3:
            cvzone.cornerRect(img, (x1, y1, w, h), t=2)
            cvzone.putTextRect(img, f'{class_labels[cls]} {conf}', (x1, y1 - 10), scale=0.8, thickness=1, colorR=(255, 0, 0))

Display the Image

Finally, we will display the processed image using OpenCV’s ‘imshow‘ method. The window will close when the ‘q‘ button is pressed.

# Display the image with detections
cv2.imshow("Image", img)

# Close window when 'q' button is pressed
while True:
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

cv2.destroyAllWindows()
cv2.waitKey(1)

Output

Output 1

Output 2

Output 3

Output 4

The Program: Detecting Potholes in a Video

The previous Python program detects potholes in an image. Now, we’ll create a different program for live pothole detection in a video. This program can also be used with a live camera to detect potholes on the road in real-time.

This program closely resembles the previous one, but here, we’ll use the ‘cv2.VideoCapture()‘ function to capture video frames and a while loop to process them continuously.

Here is the program:

import cv2
import math
import cvzone
from ultralytics import YOLO

# Initialize video capture
video_path = "Media/Potholes.mp4"
cap = cv2.VideoCapture(video_path)

# Load YOLO model with custom weights
model = YOLO("Weights/best.pt")

# Define class names
classNames = ['Pothole']

while True:
    success, img = cap.read()
    results = model(img, stream=True)
    for r in results:
        boxes = r.boxes
        for box in boxes:
            x1, y1, x2, y2 = box.xyxy[0]
            x1, y1, x2, y2 = int(x1), int(y1), int(x2), int(y2)

            w, h = x2 - x1, y2 - y1
            
            conf = math.ceil((box.conf[0] * 100)) / 100
            cls = int(box.cls[0])
            if conf > 0.4:
                cvzone.cornerRect(img, (x1, y1, w, h), t=2)
                cvzone.putTextRect(img, f'{classNames[cls]} {conf}', (max(0, x1), max(35, y1)), scale=1, thickness=1)

    cv2.imshow("Image", img)
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

Output

Summary

In this tutorial, we create a pothole detection system on roads using Python and the YOLO library (YOLOv8). From setting up the environment to training a custom YOLO model with a pothole detection dataset, we’ve covered each step in detail. We built two Python scripts here: one for detecting potholes in images and another for live detection in videos or via webcam.

This project could be beneficial in automating road inspections, saving time, and improving safety.

Recommended Article: Creating a Bike Helmet Detection Project in Python using YOLO

For any query related to this project, reach out to me at contact@pyseek.com.

Happy Coding.