Skip to content

Smart India Hackathon 2022 · DRDO

Drowning Detection

Computer-vision system that detects drowning in real time. Won SIH 2022. Published in IEEE.

Role
Team lead — Rank 1 nationally
Period
2022

#1

National rank, Smart India Hackathon 2022

The problem statement

DRDO’s brief: detect a person drowning, in real time, from video — early enough that the detection is useful rather than forensic.

Drowning does not look like drowning. It is quiet, mostly vertical, and lasts under a minute. The naive approach — flag people who stop moving — produces a false positive for everyone floating on their back.

The approach

We modelled the instinctive drowning response rather than “absence of movement”: the characteristic vertical body position, arms pressing laterally against the water, no supporting kick, and head tilted back at the waterline. Those are separable in pose data in a way that “not swimming” is not.

The system tracked pose across frames and scored the trajectory, not the frame — which is what kept the false-positive rate low enough to be trusted.

The outcome

Rank 1 nationally at Smart India Hackathon 2022, out of the full national field, on the DRDO problem statement.

The work was subsequently published in IEEE.