Ask what “billboard augmented reality” means and you will get two answers that share almost no engineering. One is a forced-perspective 3D video played on a curved LED screen, designed to look correct from a single fixed camera position — there is no augmented reality in it at all, only camera geometry and content. The other is genuine AR: a phone or headset overlaying content anchored to a real billboard in the viewer’s own camera feed. Both get sold as the same product. Only one of them is a computer-vision problem.
The divergence point is a single question: must the effect work for a passer-by holding their own device? Answer no, and you are commissioning a content and camera-geometry job. Answer yes, and you have committed to detection, tracking, occlusion handling and a latency budget that survives daylight, motion and mid-tier handsets.
What separates a 3D LED billboard from a true AR billboard?
| Dimension | Forced-perspective 3D LED | Phone-based AR overlay |
|---|---|---|
| Where the illusion lives | On the screen, correct from one fixed viewpoint | In the viewer’s camera feed, correct from any viewpoint |
| Core discipline | Content production plus camera geometry | Computer vision: detection, tracking, pose estimation |
| Viewer requirement | None — they just look | Open an app or web AR session, grant camera access |
| Fails when | The audience stands off-axis | Tracking drifts, light changes, the handset throttles |
| Primary output | A recorded clip for earned media | An interactive session with a funnel |
Neither is better. They answer different briefs, and the mistake we see most often is budgeting for the first while promising the second.
The engineering constraints that decide the outcome
Anchoring content to a physical billboard needs marker or plane tracking, stable pose estimation across the session, and a rendering budget that holds on hardware you do not control. Outdoor conditions make each of those harder: strong directional sunlight and night-time LED glare both degrade feature detection, viewing distance shrinks the billboard to a small, low-texture region of the frame, and a walking viewer supplies constant motion blur. This is the same detection-and-tracking stack described in our work on computer vision systems, applied to an uncooperative environment.
Three numbers decide whether the creative brief is reachable, and they should be named before anyone commissions artwork: tracking drift in centimetres over a 30-second session, frame rate sustained on a mid-tier handset rather than a flagship, and time-to-first-render after the camera opens. That last one is where most funnels leak — every second before the overlay appears is a viewer deciding to put the phone away.
Metrics beyond the view count
AR out-of-home teams track activation rate (the share of viewers who actually open the experience), dwell time inside it, and completion rate to the call-to-action, with earned-media impressions from the recorded clip as a separate line. A forced-perspective LED spot can only report the last of those. If the brief needs a funnel, the screen-native illusion is the wrong instrument — and if the brief only needs a shareable clip, phone-based AR adds a tracking pipeline and an access step that buy nothing.
So the honest first question is not which AR platform to use. It is whether a stranger on the pavement, in bright sun, holding a three-year-old phone, will ever see the effect at all — and what the campaign is prepared to measure if they do.
