I have almost exactly the same setup, except using the intel gpu for accelerated inference with an OpenVINO Yolo model and HA for notifications. Super reliable.