Problem
Challenge number 2 - Megatrends Extreme urbanization causes severe infrastructure damage - especially on the roads. Potholes aren't just a human problem, since the start of development in autonomous vehicles, we started to notice that driving won't be the only issue. We can drive however we want, but if the roads damage our vehicles on every step, we will not arrive far.
What it does
We built Trafficone. An AI dataset, which collects information about road damage just from images. Therefore a simple dashcam video of a road can give us the accurate information to determine the location and severity of damage. This will help drivers, who don't want their vehicles damaged. Local officials, whom need to know where the infrastructure issues are to fix them and in the future even autonomous vehicles, which will learn to predict where a pothole is forming and avoid that area.
How we built it
We created a MySQL database that contains information about roads, their exact points - so they can be drawn on the map and condition. We also created a laravel + react visualization example, a 2d map that has roads coloured based on their condition and enables any user to submit just a single photo, which then gets analyzed by AI and automatically creates a appropriate report. We used a very simple UI design with prebuilt components. The AI used is chatgpt-4o since it offers the most accuracy, but later it could be swapped for a custom AI model fed from the information we gather in a few months.
Challenges we ran into
We did not know how to work with maps in react. Had to learn a javascript library, then openstreetmap API to get accurate road locations. We also did not have any experience with OpenAI prompts, we had to create a custom prompt which would enable ChatGPT to return the required format and analyse the image properly, even knowing if there is actually a road or it's a random photo.
Accomplishments that we're proud of
We learnt a lot about AI and vectors on maps. Also we were able to build a original AI solution from scratch in just 24 hours.
What's next for Ostrov-02
In the future, images could be gathered completely autonomously, without any user input. Then we could also expand the scope to graffiti and broken lights - further solving megatrends. And in the very distant future, autonomous road repair robots could use our AI + data to prioritize areas and locate road damage.
Log in or sign up for Devpost to join the conversation.