City of Oakland / CS 2535: Professional Practicum
Illegal dumping response in Oakland is dependent entirely on civilian 311 reports, making detection slow, reactive, and unevenly distributed across neighborhoods. In partnership with the City of Oakland through a professional practicum course at Northeastern University, our team prototyped and built two image detection models to automatically flag dumping incidents from existing city camera feeds in both static and dynamic contexts, laying the groundwork for consistent, citywide coverage.
Conducted stakeholder interviews with our PI and City of Oakland administrators to scope the city's enforcement needs. Researched existing computer vision deployments in municipal sanitation contexts to benchmark approaches and outcomes. Analyzed Oakland's open 311 dataset to build a geographic heatmap tool for visualizing dumping patterns citywide.
Navigated surveillance media privacy restrictions by sourcing training data from comparable urban environments rather than live Oakland feeds. Curated images spanning varied conditions alongside a set of negative samples to reduce false positives.
Designed and implemented a transfer layer between the manual input frontend and the object detection backend. Serialized user-submitted metadata and media into detection event objects routed by type (image or video) to the appropriate model, then re-serialized the updated result for display on the output page.
.png){kind=icon}
Developed presentation materials across four months of the project, adapting format and detail level for PI check-ins, peer reviews, and mentor sessions. Designed a research posterboard for the end-of-semester showcase, communicating technical work to a broad audience spanning students, faculty, and City of Oakland stakeholders.
Small Object Detection Model Prototypes
Training Media Curated
Oakland Research Showcase Overall
