Real-Time Person Detection & Re-ID
Detect and re-identify people from live camera streams, on screen in real time.
- Role
- ML / backend engineer
- Year
- 2021
- Timeline
- Research + delivery
This is an NDA-signed client project, so it isn’t linked publicly. I’ll walk you through it live on a call and answer anything about how it was built.
A site needed to detect people from RTSP camera feeds, show results live, and recognise whether a person had been seen before on a return visit.
Video is unforgiving: the pipeline had to keep up with live streams, distinguish known from unknown people, and stay reliable enough to expose as a service.
Separated detection from re-identification behind an API.
Instead of One tightly coupled script per camera.
Why: Splitting the stages behind a REST API made the system observable, testable and reusable across feeds.
Decoupled inference from camera capture as a streaming pipeline.
Instead of Running detection inline on every captured frame.
Why: Live video never waits; decoupling let the pipeline skip frames under load and stay in sync with the stream instead of falling behind.
Matched people on lightweight appearance embeddings, not raw image crops.
Instead of Comparing each new detection against stored images directly.
Why: Embeddings turned 'have I seen this person before?' into a fast vector comparison and kept the identity store small enough to scan in real time.
- RTSP stream ingestion & detection
- Known/unknown labelling with re-identification
- Live results UI
- End-to-end pipeline behind a REST API
Hemant Manglani
Ahmedabad, India
Have a product that has to work? Let’s talk.
A 20-minute call, no pitch. We work out whether this is a real problem worth solving, and if I’m not the right person, I’ll tell you.
Every message answered within one business day