Computer Vision Analytics
People counting and dwell time for operational metrics
Operations needed to understand people flow in physical spaces to optimize staffing and service scale.
- 01Detecting and tracking people with acceptable accuracy on limited hardware
- 02Calculating dwell time even with partial camera occlusion
- 03Aggregating real-time metrics without overloading the database
YOLO for detection + IoU-based tracker
YOLO offers adequate speed for real-time processing. IoU-based tracker is simple and robust enough for low-occlusion scenarios.
In-memory aggregation with periodic flush
Writing to the database every frame would be impractical. Accumulates in memory and persists in 1-minute windows.
Data loss on crash between flushes. Acceptable given that operational metrics have tolerance for point-in-time data loss.
OpenCV + YOLO pipeline running in a dedicated process. FastAPI exposing aggregated metrics. Dashboard with flow charts by hour.
Computer vision pipeline for people counting and dwell time analysis generating real-time operational metrics.
- ›Calibrating the entry/exit counting line requires on-site testing in the physical space
- ›Lighting varies greatly throughout the day — histogram normalization improves detection consistency