Grablo Reflex User Guide

All condition choices, by analysis

Analysis	Condition choices
Object Detection	When detected · When not detected · Specific object · Except a specific object · Movement detected · No movement
Face Recognition	When known · When unknown · Any face · Specific person · Except a specific person
License Plate	Registered plate · Unregistered plate · Any plate · Specific plate · Except a specific plate · Specific alias
Fire Detection	When detected · When not detected · Any state
QR / Barcode	When detected · When not detected · Specific code · Code contains · Not this code
OCR	When detected · When not detected · Text contains · Text exactly matches · Text does not contain
Custom Classification	Specific class · Any class · Except a specific class
Custom Anomaly	When anomaly · When normal · Any state
Sound Classification	When this sound · When not this sound

Any state fires regardless of the evaluation result. Combined with a schedule, it records a daily result independent of the outcome.

For OCR, Text contains is listed first by design. Exact matching on OCR output fails frequently due to characters such as O and 0 and stray spacing.

Detection settings and defaults

Setting	Default	Range
Confidence (object)	50 %	0 to 100
Zone Trigger Point	Bottom Center	Bottom Center (Ground Point), Box Center, Overlap Ratio
Overlap Ratio	30 %	1 to 100
Threshold (face)	70 %	0 to 100
Min Face Area	6,400 px²	0 means no limit
Min Confidence (plate)	50 %	0 to 100
Detection Threshold (fire)	30 %	10 to 100
Min Confidence (OCR)	50 %	0 to 100
Min Confidence (classification)	80 %	No limit
Sensitivity (anomaly)	200 %	Training sets the boundary at 100. Lower is more sensitive (catches defects earlier), higher is more permissive
Detection FPS	5 (anomaly 1)	0 to 30, 0 means every frame
Code Type	All	QR, EAN, UPC, Code-128, Code-39, Code-93, Codabar, ITF, DataBar, DataBar-Expanded
OCR Language	English	11 languages including Korean, Japanese, Chinese
Region (plate)	Latin	Latin (EU, US, South America) or Korea

Min Face Area excludes faces captured smaller than the given area, which in practice filters out people too far away to identify reliably. The default of 6,400 corresponds to an 80 by 80 face. Raise it toward 12,544 to recognize only close, high-confidence matches.

Sensitivity requires attention to its direction. Training sets the normal/fault boundary at 100, and this value moves that boundary to the given percentage. The default of 200 places the boundary twice as far out, so ordinary variation is not judged faulty. Lowering the value pulls the boundary in, making detection more sensitive (defects are caught earlier); raising it makes detection more permissive. Raise it when lighting or background changes cause false alarms; lower it when actual defects are missed.

Glossary

Project	One complete configuration unit: cameras, connections, rules, and the log. A controller runs one at a time.
Controller	The machine running the Grablo controller software. It performs analysis and stores photos and video.
Rule	Conditions combined with AND, plus actions that run in order when all conditions are met.
Condition	One evaluation that must hold. Evaluated continuously unless it is a sub condition.
Sub condition	A condition evaluated once, at the instant the others are met, to decide whether the actions run.
Action	One task executed when a rule fires.
Analysis	One AI model watching one camera. Owned by the camera and shared by rules.
Hold time	The minimum duration a detection must last.
Detection zone	An area of the frame that filters results after detection.
Region of interest	An area of the frame given to the AI as input, before detection.
Record	One entry written to the log by a Logging action, with its photo and any video.
Enrollment	Registering a face under a name so recognition can identify that person.

Reference