Applied computer vision thesis system

Undergraduate capstone turned peer-reviewed publication: detect physical distancing cues in a web-based detector aimed at Indonesian deployment constraints.

Problem

Public-health messaging needed tooling that could run closer to users than a batch offline study, without pretending a research prototype was production surveillance.

Shape

• Object detection model trained and evaluated for the distancing use case.
• Web-facing integration path — the friction between model metrics and something a browser can host.
• Documentation and thesis narrative that later became a peer-reviewed journal article (see Credentials).
• Reference implementation: github.com/anandakevin/bachelor-thesis-social-distance-detector

What visitors should take away

Early evidence of end-to-end thinking: data, model, interface, and honesty about limits — a thread that later shows up in mobile on-device ML and platform work, with better engineering discipline.