I am building the camera-ingestion layer of a wider retail analytics platform and need it to run unattended on a Linux edge box. The service must pull live RTSP streams from multiple IP cameras, keep them alive, and surface decoded frames to the rest of the pipeline. Core requirements • Connect to any standard RTSP source and keep the video flowing in real time. • If a stream drops, the service should log the event and raise a notification; a supervising process will decide what to do next. • Frames need to be sampled at a fixed FPS that I can set per instance (e.g., 5 fps or 10 fps) and exposed through a lightweight local API so downstream AI workers can pick them up instantly. • Everything has to be written in Python with OpenCV (or GStreamer if you strongly prefer) and be easy to deploy on Ubuntu using systemd or Docker. Milestone-based delivery 1. Clean RTSP connector that plays reliably on my test Hikvision and Axis cameras. 2. Stability layer with logging + disconnection notifications. 3. Frame extractor and local API endpoint ready for integration. I will provide camera credentials and a staging box for final acceptance. The job is complete when the service runs for 24 h without manual intervention and the API consistently returns fresh frames at the configured rate.