AI-Based Road Inspection System

Local Chapter United Arab Emirates Chapter

Coordinated by,

Status: Completed

Project Duration: 15 Oct 2022 - 15 Nov 2022

Open Source resources available from this project

Project background.

The road becomes rough, bumpy, and dangerous for cars as a result of several driver violations, such as drifting, using expired tires, and driving at a fast speed. Heavy traffic on roads causes their surfaces to degrade every day which will affect the comfort of the driver as well as economic efficiency. Employing machine learning to research and evaluate many kinds of road issues and automatically spot any abnormalities on the road is an easier and more time-consuming way.

The UAE Vision 2021 National Agenda wants the country to have the best safety record in the entire world. As a result, it aims to improve its citizens’ sense of security and take the lead in the fields of safety on the roads, disaster preparedness, and security. The National Agenda also emphasizes the significance of having a just and responsive legal system that protects the rights of people and businesses and makes the UAE’s judicial system one of the most effective in the world. Furthermore, the UAE’s federal traffic law was changed on July 1, 2017. As part of Vision 2021, the new regulations seek to better secure the lives of road users by lowering the number of road deaths from 6 per 100,000 to 3 per 100,000.

UAE’s government conducts routine inspections to guarantee the security of all users of the roads. Such a process takes a long time and requires a lot of work. The procedure of road inspection must therefore be automated.

The problem.

Current practices of performing road inspections are time-consuming and labour-intensive. Road surfaces degrade on a daily basis as a result of the heavy traffic on them. This will not only impact the driver’s comfort but will also impact economic efficiency. To maintain roads as efficiently as possible, municipalities perform regular inspections. The aim of the project is to use machine learning to study and analyze different types of road defects and to automatically detect any road abnormalities. We will design, build and test an inspection system for this purpose. The system is equipped with a camera to collect video streams from different roads with and without defects. Then, the captured data will be analyzed using the Matlab machine learning toolbox to train and test the network. Finally, the system will provide recommended actions for the municipality related to actions required to fix/correct the road defects. The approach is divided into 3 main tasks: Data acquisition, Data Training/Testing, and Dashboard Building and Testing.

– Data acquisition stage: In this stage, we will use HD cameras to capture live videos of different road defects and issues. We will also collect both images from standard datasets, images from real roads, and live video recordings.
– Data Training/Testing: Collecting and labeling roads are often tedious and many times require expert knowledge. Therefore, we decided to use transfer learning to address challenges related to the scarcity of data and lack of human labels. Matlab machine learning toolbox will be used to classify road defects. We will also use standard image processing techniques to highlight areas and guide the inspection process
Dashboard Building and Testing: The dashboard or the graphical interface will visualize the defects and the recommendation to the municipality.

Project goals.

The goal of this work is to investigate the ability of various machine learning classifiers to detect road defects with the highest possible accuracy, as well as to build a Dashboard to visualize detected road defects. This project aims to: - Automate the inspection process to reduce time and effort for better efficiency - Collect as much data as possible about UAE roads (Cracks, Patching, Rutting, and deformation) - Apply machine learning algorithms and image processing techniques to detect various road defects. - Classify images using a pre-trained deep convolutional neural network (e.g., GoogLeNet, squeeznetm...etc.). - Visualize road defects in real time using dashboards/graphical user interface - Reporting the machine learning approach

Project plan.

  • Week 1

    W1.1 Research
    W1.2 Data Collection
    W1.3 Data pre-processing

  • Week 2

    W2.1 Data pre-processing/
    W2.2 Prepare/ Explore ML approaches

  • Week 3

    W3.1 Explore the pre-trained network and ML models W3.2 CNN training and validation 3.3 Performing transfer learning

  • Week 4

    4.1 System testing and accuracy reporting 4.2 Building Dashboard to visualize the output/ Deployment 4.3 Generate final report and recommendation/ Evaluate Model accuracy

  • Week 5

    4.1 System testing and accuracy reporting 4.2 Building Dashboard to visualize the output/ Deployment 4.3 Generate final report and recommendation/ Evaluate Model accuracy

Learning outcomes.

During this project, participants will be mainly able to:
– Perform data collection and pre-processing for road images
– Investigate the current pre-trained convolutional neural network, the Deep Learning Toolbox, and transfer learning
– Apply several machine learning models to classify road images – Build a Dashboard to visualize detected road defects.

Share project on: