An AI model developed in partnership with Charles Darwin University and Civiltech Solutions has the capacity to revolutionise road safety and infrastructure maintenance in the Northern Territory.
In partnership with national civil engineering provider Civiltech Solutions, a team of Charles Darwin University (CDU) postgraduate students has developed innovative technology which accurately detects and classifies road signs.
Over the course of the three-month semester, four students utilised advanced computer vision techniques to accurately detect and classify road signs in a variety of challenging conditions, including low light, adverse weather, and Darwin’s various urban landscapes.
Cong Do Le, Khai Quang Thang, Van Phuc Vinh Ho and Buu Dang Phan were tasked with the project, which was completed as a thesis project for their Master of Data Science course at CDU.
The students trained the AI model on extensive datasets of road signs in the Northern Territory, with Civiltech Solutions providing the images.
According to Leigh Carnall, Founder and Chief Executive Officer, Civiltech Solutions, the AI-powered solution offers a range of benefits.
“The model can aid in road planning by optimising sign placement and road layouts,” says Carnall.
“More importantly, the project allowed students to gain valuable real-world experience and solve real-world problems.”
Buu Dang Phan says it was an amazing opportunity to have real images to work with for this project.
“We began by receiving videos from Carnall and Civiltech Solutions which were recorded with a camera mounted on a vehicle that drove around Darwin,” says Phan.
From the 43 videos received, Phan and his three fellow team members began to cut down the video into 60,000 images and select the ones which contained road signs.
From that huge selection, the team ended up with 3000 images. However, Phan says this was too small a dataset, so they chose to use several data augmentation techniques to get more.
“We used techniques such as image rotation, scaling, and colour adjustments to further enhance the model’s ability to handle real-world variations,” says Phan.
“As a result, we accumulated 8000 images in total. After running it through many different models, we were happy with one that achieved more than 90 per cent accuracy.
“This means when you use the model, it will detect road signs and recognise them in real time with a high probability.”
The students leveraged the state-of-the-art YOLO (You Only Look Once) AI model version eight that is known for efficiently detecting objects from real-time video feeds.
They carefully curated thousands of high-quality road sign images from video feeds to achieve highly accurate image detection results through the AI model, which Phan says took a lot of hard work.
“We carefully selected the frames to make sure they all had traffic signs and we did this before using the YOLOv8 algorithm to detect and recognise the objects,” says Phan.
The project was monitored closely by Charles Darwin University’s academic supervisors Dr Yakub Sebastian, Dr Thuseethan Selvarajah and Dr Cat Kutay, who say an AI model is crucial for streamlining infrastructure maintenance.
Related stories:
- Civiltech Solutions: Data done right
- Civiltech: More, with less
- Civiltech Solutions: A key connection
“It is very important to be able to distinguish road signs from any other objects within the image,” says Sebastian.
“Once mastered, the images can be extended to a number of applications including road sign maintenance work. By identifying damaged or missing signs, this model will allow for timely repairs and replacements.”
Sebastian says road signs are often defaced or vandalised in Darwin, which is why he believes this AI model will be well received by the community.
“We can use the AI model to detect those discrepancies and flag that they need repairs or need replacing,” he says. “When deployed on individual vehicles, it may be used to enhance road safety by accurately detecting and recognising signs, preventing accidents caused by missed or misinterpreted signals.”
Although the model is currently programmed for use in Darwin, Phan says the program can easily be transferred between states.
“AI is very versatile depending on the data that goes into the model,” says Phan.
“If you feed it with a different set of road sign images from other states, it will be trained to recognise those road signs that might slightly differ from the Northern Territory imagery.
“With adequate training on varying conditions and signs, the model could work in any location in Australia.”
Unique learning
With the aim to integrate data science with engineers, Carnall knew the project would not only benefit the students partaking, but also the wider Darwin community.
“We wanted to present how data scientists can solve real-world problems and help other professions to solve those problems,” says Carnall.
“It is important that civil engineering technicians are exposed to data science, and likewise, data science technicians are exposed to real world problems.”
Carnall believes civil engineers should have an understanding of how data scientists operate, leading to a harmonious collaboration.
“We need to work with data scientists to get the best outcomes for our roads,” says Carnall. “We need to start partnering our civil engineers and the data scientists together to produce the best outcomes that are right and accurate for us.
“That was the powerful reason for engaging with the university and offering our system.”
Sebastian says the detection of road signage is just the first step in this process and is looking forward to seeing how the model expands.
“Certain things can only be achieved once you know how to detect road signs and differ them from other objects,” says Sebastian.
“This project was an important step for us and for the civil engineering industry.”
Carnall says it was an exciting first year for the collaboration, which CDU stepped up to with ease.
“We are looking forward to expanding it next year, so more CDU data science students can be exposed to real world engineering problems while exposing civil engineers to data science problems.”
This article was originally published in the February edition of our magazine. To read the magazine, click here.
