Peak 03 · 3,600 m
DDOT Computer Vision Evaluation
Benchmarked open-source CV models against a vendor for DC’s production traffic pipeline.
The problem
The DC Department of Transportation runs a production ML pipeline for micromobility detection and traffic counting that relies on a third-party vendor. The question: can open-source models match it?
The approach
Build a fair, repeatable way to compare models on DDOT’s own footage and tasks, then use the results to guide where to invest.
What I built
An evaluation framework that benchmarks open-source computer vision models against the vendor for micromobility detection and traffic counting, plus improvements to license-plate-recognition (ALPR) model accuracy.
Impact
- CVmodel benchmarking
- ALPRaccuracy improved
- Prodpipeline support