Important: This is an Early Access feature, available to selected sites behind a feature flag. Behaviour and configuration may still change during this phase.
What it does
"Analyse picture using AI" is a Triggers & Actions action that reads the picture captured earlier in your flow (for example by an IP camera), extracts a truck or trailer identifier from the image, and writes the recognised value into a visitor custom field. The typical use is verification — comparing the identifier a driver entered at registration with the one observed on the camera image. Because the AI value is written to a separate field, the original driver-entered value stays available for comparison.
What it can extract
The action recognises four identifier types. Each maps to one visitor custom field, identified by its exact technical name.
Identifier | Visitor custom field | Field type | Format the AI looks for |
Trailer number |
| Text | Painted trailer numbers, usually 3–8 digits with an optional short letter prefix (e.g. TRL 4421). |
Truck license plate |
| License plate | EU-format plates, uppercase, separators preserved (e.g. AB-123-CD). |
Trailer license plate |
| License plate | EU-format plates, as above. |
Container number |
| Shipping Container ID | ISO 6346 codes: 4 uppercase letters + 7 digits (e.g. MSCU1234567). |
Warning: These field names must match exactly. Custom aliases or tenant-specific field names are not supported during Early Access.
Before you start
Make sure the following are in place:
The target visitor custom fields, created beforehand with the exact technical names above. We recommend dedicated
[AI]fields so the driver-entered value stays untouched.An earlier action in that flow that produces a picture, such as an IP camera picture action — the action analyses the last image produced for the visitor.
How to set it up
In Triggers & Actions, edit the visitor flow that already captures a picture.
Add the action Analyse picture using AI after the picture-producing action.
Add one or more of the supported visitor custom fields to the action. Each field you add tells the AI which identifier to extract and where to write it.
To validate the result, add a following condition that compares the
[AI]field with the driver-entered field, using the existing visitor-field-to-visitor-field comparison.
Note: Standard field rules still apply — a field must be available for the profile, writable, and editable, and can be mapped only once. The AI never bypasses normal permissions.
What happens when the action runs
The action runs one analysis on the selected image and returns a result per mapped field:
Successful extraction overwrites the current value of the target field with a single, normalised value (validated against the field type first).
No writable result: the existing field value is left unchanged. The action never writes an empty or guessed value.
Partial success is normal: each field is independent, so one field can be written while another is recorded as a non-write.
When a field is not written, the outcome is recorded as one of: not found, low confidence, validation failed, or ambiguous. Values are written through the normal visitor field update path, so permissions are enforced and a visitor history line is added for each successful update.
Getting good results
Extraction quality depends heavily on the image, so aim for pictures where the identifier is clearly visible and legible. Poor lighting, rain, fog, glare, sharp camera angles, dirty or damaged labels, or several trailers in one frame typically lead to a "not found" or "ambiguous" outcome rather than a wrong value.
Not included in Early Access
The following are out of scope for now and may arrive in a later version:
Other attributes such as trailer type, transporter name, damage, or seals.
Writing to asset fields, or comparing visitor fields against asset fields.
Customer-facing configuration of the AI model, provider, or confidence threshold — these are managed by Peripass.
Feedback
Your input shapes where this feature goes next, so during Early Access please share what works and what doesn't (extraction accuracy, unclear outcomes, and setup friction) with your Peripass contact. We review Early Access performance and feedback continuously.