Skip to content

Storefront analytics on the edge

Edge AI built for the ten feet outside your door

Sample view
Your sidewalkFar side · excluded1.8 m2.9 m9.6 m

Built for the ten feet in front of a store:
the TARS video analysis engine

TARS Edge

Video frames sent off the device, by design

0frames

A design rule, not a track record: video is processed on the device and discarded.

Video ends inside the camera

A chip in the camera counts people on the spot and discards each frame immediately. Only hourly numbers ever leave.

In plain termsLike a clerk who never remembers anyone's face and only writes down how many people came in today.

FrameChipDeleted12 · 18 · 9Numbers only

Your sidewalk, not the far side

From a single camera, TARS estimates how far away each person is (monocular depth estimation) and counts only people within ten feet of your door. Because it measures distance itself, installers skip most of the tape-measure work.

In plain termsClose one eye and you can still tell a near cup from a far one. We judge distance the same way — from a person's size and where their feet land.

Excluded3 m

Every number comes with a range

Not just "420 an hour," but "90% interval 380–465," so you know how sure we are. When we say 90%, we tune the ranges so the real value lands inside nine times out of ten (calibration).

In plain termsLike a forecast that says "70% chance of rain" instead of just "rain."

Actual9 out of 10 inside the range
Sample data

Two weeks in, a full year out

For site surveys, we combine a 1–2 week sample with public population data, weather, weekdays and holidays to estimate a full year of foot traffic. The further from the measured weeks, the wider the range.

In plain termsLike estimating a year of household spending from two weeks of receipts. Times that run differently, like holidays and vacation season, are worked out separately using the calendar and the weather.

MeasuredJanAprJulOctDec
Sample data

Technology
makes the sidewalk
measurable.

Feature

Three technologies that turn the sidewalk into numbers

Detection & tracking

Detection & Tracking

Finds people in each frame and follows them briefly within the view. It never checks whether someone has been there before.

Monocular depth

Monocular Depth

Measures distance to each person with a single camera.

Funnel events

Funnel Events

Classifies passing, pausing and entering.

Differences

How it differs from a door counter

A door counter only counts people who walk through the door, so it can't tell you why others don't.

TARS counts who passes, who pauses and who walks in — so you can see exactly where you lose customers.

Typical door counter

[Entries only]

No idea who walked by or paused

TARS Edge

[Passed · paused · walked in]

Shows where you lose people, within ten feet of the door

Process

Measurement zones set up for each storefront

We fit a pass line, a pause zone and an entry line to each storefront. Thanks to depth estimation, there is almost no floor marking or tape measuring.

We'll publish accuracy figures once they are measured in pilot stores.

StoreEntry lineCameraPause zonePass line

The longer we measure,
the narrower the range

Ranges tighten as the measurement period grows. Until they're narrow enough, we don't rush to conclusions.

Estimated passersby per hour, by measurement period

Sample data
1 wk352–4882 wks373–4674 wks387–4538 wks397–443320420520

Dot: estimate · Line: 90% interval

News

Pilot updates

We'll share experiment notes from our pilot stores here.

Coming soon

Contact

Curious what your sidewalk is telling you?
Start with the overview.