What if a camera could watch the drip so nurses don't have to?

In hospitals, patients receiving intravenous therapy — glucose, saline, medications — depend on those drip bottles staying monitored. When one runs empty and isn't replaced promptly, it can delay treatment, cause discomfort, and in serious cases allow air into the line, which is dangerous.

This isn't about negligence. In busy wards, a single nurse may be responsible for a large number of patients simultaneously. Under that kind of load, manual IV monitoring becomes an unrealistic expectation.

The monitoring gap isn't caused by one thing. It's a combination of pressures that compound each other.

• High patient-to-nurse ratios

  Frequent manual checks become physically unsustainable when one nurse covers many beds.
• No remote visibility

  Nurses can't see drip status from the station — each check requires physically going to the bed.
• No proactive alert mechanism

  There's nothing in a standard IV setup that flags "this bottle is almost empty" before it happens.
• Variable drip rates

  Different patients have different flow rates, making it hard to predict when any given bottle will run out.
• Cognitive overload

  Nurses managing critical tasks in parallel are prone to delays on "routine" checks — not out of carelessness, but capacity.

One possible implementation could use a small camera mounted near each IV stand, paired with a computer vision model running on low-cost edge hardware. The model would estimate the fluid level in the bottle in real time, track how fast the drip is falling, and trigger an alert to the nurse's phone or ward dashboard when the level drops below a threshold.

A conceptual pipeline might look like this:

There are at least two broad approaches worth exploring:

A possible tech stack to explore (not finalised):

If this approach works well in practice, the potential benefits extend beyond just knowing when a bottle is empty.

• Uninterrupted patient care

  Alerts arrive before the bottle is empty — giving nurses time to respond rather than react.
• Nurses freed for higher-value work

  Removing repetitive visual checks could meaningfully reduce cognitive load across a shift.
• Low-cost deployment potential

  The hardware (camera + single-board computer) could be inexpensive enough for resource-constrained hospitals.
• Data as a side benefit

  Every drip event logged creates a record — useful for ward management, staffing, and resource planning.

I want to be upfront: this is an early-stage idea, not a working system. There are real technical and practical challenges ahead.

• Transparent bottles — fluid levels in clear plastic or glass are harder for a model to detect reliably than in coloured containers.
• Lighting variation — hospital environments have inconsistent lighting: fluorescent overhead, natural light from windows, dim night settings. The model needs to handle all of these.
• Bottle shape diversity — IV bags and bottles come in many shapes and sizes across manufacturers. A model trained on one type may not generalise well.
• False alarms — too many incorrect alerts quickly erodes nurse trust in the system. Threshold tuning and false-positive rate matter a lot here.
• Real-world deployment — hospital environments have regulations, hygiene requirements, and workflow constraints that a lab prototype won't surface.

Right now I'm in the exploration phase — reading related work, thinking through approaches, and sketching what a prototype might look like. The plan is to start with a simple proof-of-concept: a camera watching a water bottle at different fill levels, and a model learning to estimate how full it is.

From there, the real work begins — collecting meaningful data, handling edge cases, and testing whether this holds up outside a controlled setting.

I'll be sharing progress here as the project develops. If you've worked on similar problems — IV monitoring, medical CV, edge deployment in clinical settings — I'd genuinely like to hear from you.