Pothole detection system
A system installed on public transport that automatically detects road potholes and surface defects.
Tasks
- Train a model to detect road surface defects (potholes, cracks, bumps).
- { "Implement real-time processing": "detect defects and save their coordinates on the go." }
- Integrate the solution into a compact device based on Jetson.
- Reduce the workload on inspectors and accelerate response to road issues.
About the Project
The model analyzes a video stream from a camera in real time, marking problematic areas on a map with GPS coordinates. This speeds up road diagnostics and reduces the workload on inspectors.
Previously, road services relied on scheduled inspections and citizen reports to identify surface defects. This approach was slow and did not cover all issues.
We developed a modular system based on NVIDIA Jetson, combining a camera, GPS, and LTE modem, and trained a computer vision model to detect potholes and cracks. The modules are installed on public transport and collect data in the background during regular routes.
Results
Challenges and Solutions
- Limited Computational Resources
- Jetson required model optimization. We used approximation and weight compression, converting the model to TensorRT.
- Challenging Weather Conditions
- Rain, snow, and shadows interfered with detection. We added corresponding augmentations to training and applied adaptive image normalization.
- Unstable Internet
- Offline operation was enabled with coordinate buffering and data upload upon connection restoration.
Related Services
Artificial Intelligence
We create innovative solutions based on artificial intelligence for your business.
Read More
Outsourcing
Our team of skilled developers is ready to create websites and applications of any complexity. We build modern web solutions using cutting-edge technologies and architectural approaches.
Read MoreReady to discuss your project?
Describe your task, we will make a research and respond to you as soon as possible.