Privacy-Preserving Crowd-Density Estimator for Transit Stations
Overview
What this challenge is about.
Use a public crowd-counting dataset (e.g., ShanghaiTech or JHU-CROWD) to train a small crowd-density estimator (CSRNet or similar). Wrap it in an on-device pipeline (Python is fine for the prototype) that ingests video frames, computes a per-zone density map, writes only the aggregated numbers to disk, and discards the frame. Add an automated test that fails if any raw frame is persisted beyond 200 ms. Measure Mean Absolute Error of headcount on the dataset's test split, plus per-zone density error on a synthetic 4-zone overlay. Write the privacy review document (3 pages): data minimization, threat model, retention story, and a clear answer to 'what could go wrong if this pipeline is misconfigured?'.
The Brief
What you'll do, and what you'll demonstrate.
Ship a privacy-preserving crowd-density prototype with both a working perception model and a defensible privacy review document.
Earning criteria — what you'll demonstrate
- Train a density-map crowd estimator and evaluate it with Mean Absolute Error
- Apply data-minimization patterns to a perception pipeline (never persist raw frames)
- Write a threat-modeled privacy review for a municipal stakeholder
- Communicate scene-understanding outputs as aggregated, privacy-safe numbers
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
Career paths this builds toward
Canonical rolesAI Product Designer
Designing a perception product around hard privacy constraints and writing the reviewer-facing rationale is exactly the day-one job of an AI product designer at any civic-tech or smart-cities team.
This challenge sharpens
- privacy-by-design
- scene-understanding
- edge-deployment
Computer Vision Engineer
Training and deploying a small density estimator under edge constraints is the bread-and-butter CV-engineer work behind any 'count without recognize' product.
This challenge sharpens
- crowd-counting
- scene-understanding
- edge-deployment
AI Safety Researcher
Threat-modeling a perception pipeline and writing the misconfiguration-failure-mode section is directly relevant to AI safety work on deployed systems.
This challenge sharpens
- privacy-by-design
- evaluation
- scene-understanding
AI Product Manager
Owning the trade-off between counting accuracy and zero raw-frame retention is the kind of decision an AI PM makes on every civic deployment.
This challenge sharpens
- privacy-by-design
- edge-deployment
- evaluation