A traffic camera system that counts cars without needing the cloud

Researchers built a real-time vehicle detection system that runs entirely on a local computer, using a pretrained AI model (YOLOv11) to identify and count cars from video feeds without sending data to cloud servers. In typical conditions, the system identifies cars with 97-100% precision and counts them with 67-96% accuracy, making it deployable in cities that want traffic monitoring without infrastructure costs or data privacy exposure.

Most traffic monitoring systems either require cloud processing (expensive, slow, privacy risk) or custom hardware (capital-intensive). This paper demonstrates that off-the-shelf AI models can run on cheap local hardware and still deliver useful accuracy — which means a city doesn't need to buy specialized equipment or pay recurring cloud fees to know how many vehicles are moving through an intersection. The constraint isn't capability anymore. It's just whether municipalities decide to deploy it.