Monash DeepNeuron Pulse A DEEPNEURON PROJECT

Pulse: a camera that watches when carers can't

A camera that watches when carers can't.

WEEK 6 · PIPELINE · SHOWCASE IN 2 WEEKS
WHERE THINGS STAND

Four things worth knowing before you read further.

Pulse is a device that watches over bedridden patients in the hours between carer rounds. A CNN-LSTM pipeline reads facial pain expression over time, a separate pose model tracks movement out of bed, and planned environmental sensors will add a fuller read on patient state, so problems are caught continuously instead of at the next check.

Nobody is in the room at 3am.

Up to four hours can pass between rounds for a patient who can't call for help. We're aiming to close that gap to closer to forty-one seconds.

Two branches: a face, or a body.

In bed, a blanket hides the body, so Pulse reads the face. Out of bed, a pose model reads 17 body keypoints instead. Environmental sensors, planned next, add context from the room itself for a fuller read on patient state.

A laptop, a webcam, and a CAD file.

  • DONEFace and pose detection
  • DONECNN trained on pain expressions
  • IN PROGRESSLSTM pain score
  • IN PROGRESSLive pipeline into the carer dashboard
  • IN PROGRESSThe 8cm cube, hardware arrived for assembly

Live video in, alerts out.

Before the 7 October showcase: finish the LSTM and wire everything into one live loop. After that: environmental sensors, and testing how a clinic dataset holds up in a dark bedroom.

Read the full project

We'd rather show the project at this stage than wait for one that looks finished.