Low Frame-Rate Vehicle Counting and Tracking — Bath, UK

Bath & North East Somerset Council (B&NES) needed to gather fine-grained vehicle counts from traffic footage without installing permanent camera infrastructure. To keep their studies flexible, they deployed battery-powered cameras that can be moved between sites on demand — but running on battery power meant recording at a much lower frame rate than a typical fixed installation.

Low frame-rate footage makes vehicle tracking substantially harder: vehicles can jump large distances between consecutive frames, motion cues become unreliable, and standard tracking-by-detection approaches start to break down. This project focused on solving these tracking complexities, combining detection and identity-matching strategies designed specifically to cope with large inter-frame gaps.

The resulting pipeline was validated against a human-labelled dataset and achieved 98% counting accuracy, giving B&NES a reliable way to collect fine-grained traffic data from flexible, infrastructure-free camera deployments.