Target passers-by using computer vision at the edge
With a webcam connected to a display, Personalisation Hub can detect who is walking past and switch the campaign to match. For example, it can play a men's offer when the passer-by is male. The computer vision model runs on the display's own device at the edge; no images or sensitive data leave the store.
Before you begin
- A display with a webcam connected, registered under Displays & Devices.
- A campaign with creative for that display's playlist.
- For what the platform does and does not keep, see Does computer vision store images of shoppers?.
Add the targeting rule
- In the left navigation, select Campaigns, then select the campaign name.
- Select the Targeting tab.
- Select Add New "AND" Condition.
- In the first select, choose Visitor / Customer Data.
- In the second select, choose Gender.
- In the third select, choose equal.
- In the fourth select, choose male or female.
- To have the campaign interrupt whatever is playing as soon as a passer-by is detected, turn on Override current campaign.
- Select Save Changes, then turn on the activation toggle at the top right.
Enable Vision/AI on the display
- In the left navigation, select Displays & Devices.
- Select the display name, then select the Enabled Features tab.
- Turn on Enable Vision/AI (BETA).
- Tick Monitor Passerby & Campaign Engagement Data. Tick Record as shopfront passer-by traffic if the display faces the street.
- Leave Detection Preset on Balanced unless the device struggles (Fast) or misses short visits (Accurate).
- Select Save Changes.
Verify it worked
Stand in front of the display. As soon as the camera picks you up, the matching campaign starts; with Override current campaign on, it cuts in over the campaign that was playing. Views and passer-by demographics then appear under Insights / Analytics — see Review campaign performance in Insights / Analytics.
Troubleshooting
Nothing changes when someone walks past
Cause. Vision/AI is not enabled on that display, or the webcam is not connected to the device.
Fix. Check the Enabled Features tab shows Enable Vision/AI (BETA) ticked and saved, then check the camera on the device.
Detection is slow or misses people
Cause. The Detection Preset is too light for the traffic, or the device's CPU is saturated.
Fix. Try Accurate on capable hardware, or Fast on weak hardware, and save.
Related
Last updated 17 September 2026