sammy1am / doods2 Goto Github PK

MQTT image cropping happens sort of inline with other MQTT processing, but would be cleaner in its own class/function (class to store the original detected, decoded image in case multiple crops need to be made).

Currently each object detected is published to MQTT separately, so there's no way to tell if e.g. there were two people detected in frame, or one person detected twice over two frames.

• Adding some sort of frame serial number might allow MQTT consumers to group these detections together.
• Alternatively, detections could be grouped into a single JSON message for publishing. This makes it more difficult for consumers to determine simply questions like "was a person detected in this region" since they would have to parse out the various detected objects first. On the other hand, it makes it way easier to answer questions like "how many people are here". This would make cropped images harder to associate though unless we move to providing images as base64 encoded within JSON responses (instead of binaries in the MQTT stream), which might break Home Assistant integration.

The latter option would be a breaking change, but if I change it soon hopefully doesn't break too many peoples' things.