Authors

Akbar Abilov, Lochan Pranav M, Clement Mo

Research Question

Can cameras already mounted on Paterson's street sweepers serve as a passive, citywide detection network for illegal dumping and littering? Building on a previous capstone, this project explores what detection accuracy is achievable, how confidence thresholds should balance enforcement value against false citations, and what human-review workflow is needed before a detection supports a warning or citation.

Background

Illegal dumping is hard to catch in the act, and enforcement currently depends on inspectors being in the right place at the right time. Street sweepers already travel the city on a regular schedule, acting as a mobile camera network that goes largely unused for enforcement. Turning this existing infrastructure into an active detection tool could increase detection volume and consistency without a proportional rise in staffing costs, directly supporting the city's cleanliness ordinances and citation program.

The project extends a prior capstone, so the city gets compounding value from its research investment. Paterson is one of 700+ Keep America Beautiful (KAB) affiliates; KAB will help fund AI and compute costs and may promote the solution across its affiliated cities if it succeeds.

Methodology

The project involves:

• Working with sweeper video footage, upgraded to higher resolution as needed.
• Developing or refining computer vision models that detect trash, dumped items and dumping-in-progress activity.
• Comparing detection accuracy and false-positive rates with current versus upgraded camera resolution.
• Tuning confidence thresholds to balance enforcement value against the risk of false citations.
• Designing the human-review workflow required before a detection supports a warning or citation.
• Moving from prototype to a deployable model, rather than starting from scratch.

Deliverables

• Interactive data dashboard
• Trained computer vision / detection model, moved from prototype to a deployable version
• Model performance evaluation (accuracy, false positive/negative rates)
• Proposed warning/citation workflow integrating detections
• Technical documentation

Data Sources

• Street sweeper video footage from the City of Paterson
• Outputs and models from the prior capstone project