01Start with the trigger
Presence/staying accepts the whole image or one zone per camera. A crossing rule accepts one tripwire and its allowed A→B, B→A or both directions, not a zone or mask.
Several cameras can share conditions and recipients but keep their own scope, events and cooldown. Couch and bed means either selected class can match; it does not require both. Free-text names/descriptions explain the saved settings; they are not arbitrary LLM instructions.
02Select areas without mixing their roles
A rule zone limits where objects qualify. A motion-ignore mask affects low-level motion measurement, not YOLO pixels and not a blanket ban on alerts. A rule-exclusion area only applies when selected in that camera’s presence/staying or hardhat rule.
Objects enter zones according to the bottom-center of their detection box. Coordinates are proportional to the image. Resolution changes alone preserve geometry, but camera movement, a new crop or a changed field of view requires review.
03Use foreground priority only as an exception
An enabled foreground-priority area exempts objects whose box bottom-center enters it from rule exclusions. Objects must still match the class, trigger scope and timing. This is not distance estimation or a confidence boost; it does not override motion masks.
Crossing does not support exclusions or foreground priority. For a doorway with a distant sidewalk, a visitor’s upper body overlapping the sidewalk exclusion is not enough to exclude them if their bottom-center is outside.
04Check the result and delivery
Save and enable the rule, enable delivery, and choose configured recipients. New rules default to a 60-second cooldown per camera. This is a minimum repeat interval, not a promise to send every minute while a condition remains true.
Test a realistic pass through the area and inspect Events and delivery status. Copy creates an independent disabled draft. Removed or unavailable trigger scopes do not silently become whole-image rules.
Common questions
Does drawing a region start alerts?
No. Bind it to an enabled rule with the right objects and recipients. Whole-image presence needs no drawn region.
Should I create one rule per camera to save GPU?
Grouping mainly simplifies management. Object detection is shared per camera; each additional camera still needs processing. PPE adds a separate check.