From ecad939a62bd51cfb1fedc294a7c2140debebce2 Mon Sep 17 00:00:00 2001 From: Kai Curry Date: Mon, 9 Feb 2026 14:36:35 -0100 Subject: [PATCH] Add detail to face recognition MQTT update docs Clarify that the weighted average favors larger faces and higher-confidence detections, that unknown attempts are excluded, and document when name/score will be null/0.0. --- docs/docs/integrations/mqtt.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/docs/docs/integrations/mqtt.md b/docs/docs/integrations/mqtt.md index 535e1bb4b..5020d7955 100644 --- a/docs/docs/integrations/mqtt.md +++ b/docs/docs/integrations/mqtt.md @@ -134,6 +134,8 @@ Message published for updates to tracked object metadata, for example: #### Face Recognition Update +Published after each recognition attempt. The `id` is the tracked object's event ID, which can be used to look up the event via the API or match it to items in the UI. The `score` is a weighted average across all recognition attempts for this person object, not the score from a single attempt. The weighting favors larger faces (by pixel area, capped at 4000px) and higher-confidence detections. Attempts scored at or below `unknown_score` are excluded from the average. The `name` is the best matching person based on that weighted average, or `null` if no match. Both `name` and `score` will be `null` / `0.0` if fewer than `min_faces` attempts have been recorded or if two names are tied for the most detections. This message is published regardless of whether the score meets `recognition_threshold`. + ```json { "type": "face",