Deep Learning Based Road Cracks Detection using UAV Images

In the era of automation, automated road crack detection is an important task for the maintenance of transportation for safe driving assurance. It has always been a challenging task due to the complex pavement conditions, complexity of the background and the cost effectiveness. We propose a road cra

2025-06-28 16:26:05 - Adil Khan

Project Title

Deep Learning Based Road Cracks Detection using UAV Images

Project Area of Specialization Artificial IntelligenceProject Summary

In the era of automation, automated road crack detection is an important task for the maintenance of transportation for safe driving assurance. It has always been a challenging task due to the complex pavement conditions, complexity of the background and the cost effectiveness. We propose a road crack detection system that is based on deep learning models. The dataset of our project is of 10,000 images which is collected using Unmanned Aerial Vehicle (UAV). Also, two types of cracks i.e., alligators and potholes are proposed to be detected by using state of the art deep learning-based object detection algorithm such as You Only Look Once version-3 (YOLOv3) to localize the cracks and observe the performance. For real time implementation, we use Raspberry Pi controller.

Project Objectives

The main objectives of our project are as follows:

Project Implementation Method

Project implementation is described by the following steps:

Dataset Collection using UAV:

We collect the videos of cracked roads from the roads of Swabi. As we are doing our project on image dataset so we extract the frames from those videos to make our image dataset for further processing. This dataset is composed of 10,000 images of cracked roads in which two types of cracks are mainly focused which are alligators and potholes.

Pre-Processing of Dataset:

In this step, we have to pre-process our image dataset because these are the images which are used for training as well as testing the deep learning model. So, it should be in refined form. So, for this purpose, we cropped the unwanted area from the images and set the aspect ratio to 3:4 which gives the size of 416 x 416. All the pre-processing is done by using MS Picture Manager. After this we have to rename all the images with their respective crack class number.

Dataset Annotation:

For this purpose, we use labelimg tool for ground truth generation. In this tool we just have to draw the bounding boxes around the cracked area and assign the name of the class to the specific cracks. After this, we have to save the annotation which is in YOLO format which provides us the information of class number and the coordinates of the cracks which will be used in training process.

Training and Testing of Deep Learning Model:

In this step, we have to train our model using YOLOv3. As our dataset consists of huge amount of data so for the training purpose, we have to purchase our own GPU. We set 80% data for training purpose and 20% data for testing purpose. After training the model we got the file of weights of our trained model which are used for the testing of our model by using test data set.

Localization of Cracks by Generating their Coordinates:

After training and testing of our model, we have to do the detection on the video of the cracks to check rather it gives us the right detection or not. Also, the accuracy along with the coordinates will display in the video.

Raspberry Pi Implementation:

As we have to do the Raspberry Pi implementation, so in this step we have to do the detections on Raspberry Pi. For this purpose, we have to write the Python script to follow the exact steps which are mentioned above except training because we already have the file of weights of our trained model so we directly use them for the detection purpose. We just have to load the data for the detection and then output will also be generated using Python script.

Benefits of the Project

The benefits of our project are as follows:

Technical Details of Final Deliverable

The final deliverable of our project will include the following:

Final Deliverable of the Project HW/SW integrated systemCore Industry ITOther Industries Education Core Technology Artificial Intelligence(AI)Other Technologies Internet of Things (IoT)Sustainable Development Goals Decent Work and Economic Growth, Industry, Innovation and Infrastructure, Sustainable Cities and CommunitiesRequired Resources
Item Name Type No. of Units Per Unit Cost (in Rs) Total (in Rs)
Total in (Rs) 63850
NVIDIA Quadro K5000 4GB Equipment14000040000
Power Supply for GPU Equipment170007000
Camera Module for Pi Equipment190009000
SD Card 32 GB Equipment1700700
Card Reader Miscellaneous 1150150
Other Expenses (Printing, Posters, Documentation) Miscellaneous 170007000

More Posts