FACETRACE
REAL-TIME COMPUTER VISION
FaceTrace / Drishti
Multiple camera feeds. One live pipeline for detection, recognition and alerts.
THE IDEA
FaceTrace brings detection, identity matching and stream delivery into a live vision application. Its architecture prioritises current frames so slow inference does not turn into an ever-growing video backlog.
HOW IT WORKS
- 01Receive camera frames
- 02Select the latest frame
- 03Run detection
- 04Match face embeddings
- 05Deliver live alerts
INSIDE THE BUILD
ONNX inference supports local object and face-processing workloads.
Per-stream processing separates camera ingestion from model execution.
Embedding comparisons match faces against enrolled records.
A web interface brings streams and identity or alert information together.
In a live interface, a recent frame is often more useful than processing every stale frame.
THE TRADE-OFFS
Recognition quality depends on camera conditions and enrolled data. CPU inference, occlusion and concurrent stream counts constrain throughput; alerts still require human judgment.
BUILT WITH
- Python
- ONNX
- FastAPI
- React
- MongoDB