Skip to content
appunik adDeviceBook a demo

How it works

From a camera on your screen to a report your advertisers trust

adDevice runs entirely on a small computer behind the screen. It plays your ads, measures attention and sends only anonymous numbers to the dashboard.

The pipeline

Images go in. Only numbers come out.

  1. Step 1

    Camera

    Sees the people in front of the screen. Frames stay in memory.

  2. Step 2

    On-device AI

    Finds faces, works out distance and line of sight, then wipes the frame.

  3. Step 3

    Anonymous numbers

    Counts and gaze seconds per 10-second window. No images, no IDs.

  4. Step 4

    Dashboard

    Per-ad statistics, heat maps, playlists and PDF reports.

The frame is wiped the moment the analysis is done. The privacy boundary sits between steps 2 and 3: nothing image-like crosses it. Read the privacy details.

How looking is detected

Line of sight, not just a face in the frame

For every face the device works out where the head is (distance and direction from the camera) and which way it points: head pose, plus the eyes when the face is close enough to read the iris. If that line of sight meets the screen, the person is looking.

  • No calibration for viewers. People just walk past. An optional one-time alignment at install corrects small mounting errors.
  • Range. Looking and dwell time work for anyone the camera can see within about 4 m. Off to the side, accuracy is around the head-pose error (about 8°).
  • Blinks and missed frames. A glance is kept together across blinks and dropped frames, so one look is one look.
  • Dim halls and backlight. Faces are contrast-boosted and dark frames brightened before detection. For very dark spots we can fit an infrared light.

Three levels of attention, each one includes the next

  1. 1

    Glances

    Counted when: Their line of sight touched the screen, even for a moment while walking past.

    Your true reach: how many people the screen caught, not how many walked by.

  2. 2

    Viewers

    Counted when: They looked for at least half a second.

    People who actually saw the ad, not just the screen.

  3. 3

    Engaged viewers

    Counted when: They kept looking for 2 seconds or more.

    The audience that read your message. The number creative teams optimise for.

Heat map and areas of interest

See which part of the ad people read

Gaze is collected on a 32 × 18 grid over the ad and summed per area you draw: logo, product, price, button. The heat map and areas of interest use viewers close enough to tell one part of the ad from another (on a 27" screen, within about 1.4 m). Further away, people still count as looking.

Where people looked

Gaze time per area

  • Product46%
  • Headline28%
  • Price4%
  • Buy button4%
  • Logo2%
  • Subtext1%

Areas are drawn on the ad in the dashboard. Draw one later and it is estimated from the stored heat map.

Sample data on our demo ad, for illustration.

The platform

Run the screens, not just measure them

Playlist management

Upload images or MP4/WebM videos (1920 × 1080 or 1080 × 1920), set how long each one shows, reorder, enable or disable. Screens pick up changes within about 15 seconds.

Keeps playing offline

A new playlist only goes live once every file is downloaded and checked. Without a network the screen keeps playing its cached playlist.

Areas of interest

Draw logo, product, price or button areas on the ad in the dashboard. Areas drawn later are estimated from the stored heat map.

PDF reports

Export the statistics for one ad, or every active ad in the playlist, as a PDF to share with your client or creative team.

Per-device keys

Each screen gets its own key, created in the dashboard. You can see whether every device is up to date.

Any orientation

Landscape or portrait. The device reads the screen size from the HDMI monitor and adapts the heat map grid.

Watch it work on your own screen

We bring the device to your site, mount it in about 15 minutes and show you live results on the dashboard.

Book an on-site demo