AI Pulse Labs tests face-mesh billboard tracking for ad attention

• In youraipulse's demo thread, AI Pulse Labs and liliumdev shared a rough MVP that counts pedestrian traffic, estimates head orientation, and approximates whether people actually looked at a billboard.
• AmirMushich's post framed the project as an attempt to measure real attention instead of raw impressions, with a face-mesh tracker feeding a live ad-analysis view.
• The team also said, in AmirMushich's post, that the goal is to analyze pointers while prohibiting data collection and storage, which puts privacy constraints inside the experiment itself.
• The replies added two important caveats: AmirMushich's reply about billboard owners said the prompt came from real owner questions about visibility, while AmirMushich's exploration note said the project is not being sold and still needs much more research.

You can watch the demo clip draw boxes around pedestrians and mark gaze direction in real time. AmirMushich's post is more explicit about the privacy angle, while his reply about indoor billboard owners explains the real-world question behind the build. Even the throwaway replies are useful here: youraipulse tagging Roboflow hints at the computer-vision tooling orbit, and AmirMushich's follow-up makes clear this is still a lab experiment, not a product.

The prototype is very specific. In youraipulse's thread, the team says it estimates pedestrian traffic, detects head orientation, and approximates whether passersby actually looked at the ad.

That is a tighter claim than standard out-of-home reporting. As youraipulse's post puts it, the target is attention, not impressions or traffic.

Realtime billboard attention demo

The demo video shows the system turning a street camera into an attention layer: bounding boxes on pedestrians, orientation lines, and a dashboard-style readout. AmirMushich's reply about tracking suggests the team is already thinking beyond foot traffic, including questions like passenger counts inside cars.

That matters because the measurement unit here is not just who passed the billboard. It is who plausibly had the billboard in view, using head pose as a rough proxy.

The most interesting detail in AmirMushich's post is the attempt to "prohibit the data collection/storage" while still extracting attention signals. That makes the experiment feel closer to an edge analytics layer than a surveillance archive.

The evidence does not explain the stack or retention policy in detail. What it does show is a team trying to separate aggregate attention measurement from long-term identity storage.

The replies add the missing business context. In AmirMushich's reply about billboard owners, he says they had been co-working with an indoor billboard owner who kept asking, "How many people will see my ad?" and that the team had "zero ideas" before this experiment.

They are also downplaying any immediate commercialization. AmirMushich's exploration note says they are not selling it anywhere and are building it "for the sake of exploration" for now, while AmirMushich on the MVP lab describes the group as a products and MVPs lab.