Object detection & AI backends

Object detection finds people, vehicles, animals, and packages in the frame, and is the basis for faces, license plates, and semantic search. It runs on an AI backend you choose to match your hardware.

Choosing a backend

camera.ui ships four AI backends as plugins. Pick the one for your hardware and enable it per camera in Settings → Plugins (see Set up sensors):

• CoreML. For Apple Silicon Macs; uses the GPU and Neural Engine.
• OpenVINO. For Intel CPUs and GPUs.
• ONNX. Cross-platform, on CPU or a supported GPU.
• NCNN. A lightweight backend for supported hardware. (No semantic search.)

Each backend provides object, face, and license-plate detection. CoreML, OpenVINO, and ONNX also provide CLIP for semantic search.

Models and confidence

In a backend's plugin settings you choose a model for each task (object, face, license plate, CLIP) and a confidence threshold, which is how sure the AI must be before it reports a detection. Larger models are more accurate but heavier; higher confidence means fewer false alarms but more missed detections. camera.ui downloads the models it needs automatically.

There is also a per-camera Object confidence in Settings → Detection; set it to 0 to skip object detection on that camera.

What you get

Detected objects appear on the live view as boxes, drive notifications, and become events you can browse and filter by type. They also feed face recognition, license plates, and semantic search. A heatmap overlay on the player shows where objects were detected most often over a time window.