The challenge
A system integrator executing a governmental project needed a road traffic analytics system that could identify road entities such as cars, pedestrians and bicycles, track them and analyse them further, for example by recognising a car's type, make, model and colour.
The main challenge was robustness: the system had to transfer easily across different camera types and points of view, and work in different environments.
Our solution
We developed a system that processes video streams from traffic cameras and analyses road scenes: it detects objects, estimates their properties such as trajectory and speed, identifies traffic jams and violations, and recognises car models and licence plates.
Each analytical component was built as an independent module, so the system can be customised. It detects and classifies partly occluded objects, identifies a car model's production year, supports commercial vehicles, adapts to different traffic environments and can be updated with new car models and further features such as advanced tracking or event detection.
- Detection of cars, pedestrians, bicycles, trucks, buses and motorcycles, with object tracking
- Car speed estimation and colour recognition
- Car type, brand and model recognition across 7,000+ unique models
- Licence plate recognition and validation
- Traffic intensity analysis and illegal parking analysis
- Anomaly detection, such as a pedestrian on a highway
The outcome
The system was deployed by a city municipality and runs 24/7 across hundreds of cameras, giving the city a safer environment and the ability to monitor its roads without human intervention.
