Multi-Camera Vehicle Tracking — Carmel, Indiana

EASE-MCVT is a real-time and scalable framework for understanding how vehicles move through an area covered by multiple traffic cameras. Instead of treating each camera as a separate source, it connects the views together to build a more complete picture of traffic movement.

The framework detects and tracks vehicles in each camera view, then re-identifies the same vehicle as it moves from one camera to another. This allows us to reconstruct complete vehicle journeys across the monitoring area, including the parts of a journey that are captured by different cameras. EASE-MCVT can use existing city-camera footage for offline analysis, or run on edge devices connected to cameras for real-time processing.

These vehicle trajectories can support a range of practical applications, including traffic-flow modelling, traffic prediction, transport planning, traffic analysis, traffic anomaly detection, and retrieving the movements of a vehicle of interest. The framework is designed to be scalable, making it suitable for deployments that involve multiple cameras and large areas of coverage.

Alongside EASE-MCVT, we collected and open-sourced RoundaboutHD, a dataset created to support further research and development. It enables work on object detection, object tracking, geo-mapping, vehicle re-identification, and multi-camera tracking, helping researchers and developers evaluate new methods on realistic traffic-camera data.