Real-Time CCTV Person Tracking

I need a complete computer-vision pipeline that lets me watch multiple indoor and outdoor cameras and immediately know who is where. The core goal is security surveillance, so every component has to run in real time without noticeable lag. Here is what the system must do: • Detect every person entering the frame, day or night, in varied lighting and weather. • Track each individual seamlessly across several overlapping or non-overlapping cameras (multi-object tracking). • Re-identify the same person when they reappear on a different camera, even after minutes out of view. • Expose a live stream API or dashboard that shows bounding boxes, unique IDs, and confidence scores as events unfold. I am open to the specific model stack—YOLO, Detectron2, DeepSORT, StrongSORT, or any other proven combination—as long as you can meet real-time FPS on standard GPU hardware and keep false positives low. Please factor in calibration for both indoor and outdoor lenses, simple configuration files for adding new cameras, and a straightforward way to export clips or snapshots if a tracked ID triggers an alert. Acceptance criteria: 1. End-to-end demo on at least two indoor and two outdoor cameras running simultaneously at 25 fps+. 2. Consistent ID hand-off across cameras with ≤5 % ID-switch rate on a one-minute test sequence. 3. Setup script and concise readme so I can replicate results on my own machine. If you’ve built something similar or have benchmark videos to prove speed and accuracy, that will move things forward quickly.