Automatic Road Pavement Distress Detection Through Deep Convolutional Neural Network
Road potholes and cracks are the types of road pavement distress, these factors jeopardize the safety of road and transportation efficiency. There are different types of road pavement distress like Fatigue Cracking, Block Cracking, Depression, Joint reflection cracking, Longitudinal cracking, Patchi
2025-06-28 16:30:28 - Adil Khan
Automatic Road Pavement Distress Detection Through Deep Convolutional Neural Network
Project Area of Specialization Artificial IntelligenceProject SummaryRoad potholes and cracks are the types of road pavement distress, these factors jeopardize the safety of road and transportation efficiency. There are different types of road pavement distress like Fatigue Cracking, Block Cracking, Depression, Joint reflection cracking, Longitudinal cracking, Patching, Polished Aggregate, Potholes, Raveling, Rutting, Slippage cracking, Stripping and Transverse thermal cracking. To identify these types of distresses road assessment authorities, need specialized teams of inspectors and structural engineers who manually assess road infrastructures and provide detailed reports about the detected pavement distress type. Furthermore, these are some of the main factors which effects transportation efficiency and cause of vehicle accidents. The objective of the proposed study is to develop a system for the road safety and assessment, which will detect and identify road conditions in real-time. The proposed system will identify the major six pavement distress ( Fatigue Cracking, Joint reflection cracking Potholes, Patching, Rutting, and Depression) using small IP camera mounted in front of vehicles and provide a helpful tool to road assessment authorities to assess the road pavement conditions in real time. The system will serve the vehicle drivers by notifying them about road conditions in real-time which will help them to avoid any damage and will helpful for highway authorities to assess the different road distress types. The system will be trained on manually collected dataset of road distressed images using a deep convolutional neural network for real-time detection.
Project Objectives1. To detect and identify road cracks and potholes.
2. To recognize major types of road pavement distress.
3. To develop prior intimation system about road conditions by capturing real time image.
| Our methodology of proposed system will contain four phases.
Dataset Collection Images of 6 main distresses of road in Pakistan will be collected: Fatigue Cracking, Joint reflection cracking Potholes, Patching, Rutting, and Depression Preprocessing of Data The aim of pre-processing is an improvement of the image data that suppresses and enhances some important features of images for further processing, although Rescaling, Noise Removal (DE noise), grey conversion and morphology of images are classified among pre-processing methods here since similar techniques are used. Training custom object Detection model In second step the images after preprocessing will labeled manually and create a annotations. These Annotation will contain the information of Region of interest. These annotations will split into test and train from these features vector are extract and will used to train and test the pre trained deep convolutional neural network model Single shot detection (SSD) As in figure. The learn model will be able to detect and identifies the road cracks and pot-holes. Validation of learned model. The learned model will be validate using raspberry pi and IP camera mounted in front and the LED inside the car. The LED and IP camera will have connected on same network. Workflow
|
Our methodology of proposed system will contain four phases.
-
Dataset collection
-
Preprocessing of Data.
-
Train DCNN to detect and Identifies Road Pavement Distress.
-
Validation of learned DCNN.
Dataset Collection
Images of 6 main distresses of road in Pakistan will be collected: Fatigue Cracking, Joint reflection cracking Potholes, Patching, Rutting, and Depression
Preprocessing of Data
The aim of pre-processing is an improvement of the image data that suppresses and enhances some important features of images for further processing, although Rescaling, Noise Removal (DE noise), grey conversion and morphology of images are classified among pre-processing methods here since similar techniques are used.
Training custom object Detection model
In second step the images after preprocessing will labeled manually and create a annotations. These Annotation will contain the information of Region of interest. These annotations will split into test and train from these features vector are extract and will used to train and test the pre trained deep convolutional neural network model Single shot detection (SSD) As in figure. The learn model will be able to detect and identifies the road cracks and pot-holes.
Validation of learned model.
The learned model will be validate using raspberry pi and IP camera mounted in front and the LED inside the car. The LED and IP camera will have connected on same network.
Workflow
| The system will serve the vehicle drivers by notifying them about road conditions in real-time which will help them to avoid any damage and will helpful for highway authorities to assess the different road distress types. The system will be trained on manually collected dataset of road distressed images using a deep convolutional neural network for real-time detection.
|
The system will serve the vehicle drivers by notifying them about road conditions in real-time which will help them to avoid any damage and will helpful for highway authorities to assess the different road distress types. The system will be trained on manually collected dataset of road distressed images using a deep convolutional neural network for real-time detection.
- Increase road safety
- minimize accidently damage
- help the highway authorities in road inspections
| N o | Name | Functionality | Image |
| 1 | Respberry Pi 4 8Gb RAM Quad Core CPU 1.5 Ghz | It is a microcontroller giving desktop performance as PC |
|
| 2 | GPS Tracker | Watch multiple vehicles on one screen |
|
N o
1
2
Final Deliverable of the Project HW/SW integrated systemCore Industry PetroleumOther IndustriesCore Technology Artificial Intelligence(AI)Other TechnologiesSustainable Development Goals Industry, Innovation and InfrastructureRequired Resources| The system will serve the vehicle drivers by notifying them about road conditions in real-time which will help them to avoid any damage and will helpful for highway authorities to assess the different road distress types. The system will be trained on manually collected dataset of road distressed images using a deep convolutional neural network for real-time detection.
|