Foot traffic data: what open sources actually provide
Foot traffic sounds like the most basic number in site selection, and the easiest to get. It is neither. It is a count of people crossing one line, at one place, over one period - and almost every dataset that carries that name is a sample of a sample. This page names the real sources, states what each of them covers, and says plainly what none of them covers. LocateRight no score, ranking or recommendation is published, so nothing here is dressed up to look more capable than it is.
Three different things get called foot traffic
Before comparing sources, separate the measurements. They are routinely substituted for one another, and the substitution is where site decisions go wrong.
- A physical count - people crossing a line, counted by a sensor, a camera or a person. This is what foot traffic means strictly.
- An activity estimate - a density signal over an area, derived from device telemetry. Proportional and useful for comparison, but not a headcount.
- A catchment population - the number of residents inside a boundary. It measures where people live, not who passes your door.
A high catchment population with a low count, or the reverse, is a common and important finding. Neither figure implies the other.
Five sources, and what each one actually gives you
These are the sources we could verify by reading their own documentation. Each link goes to the publisher, not to a summary.
City of Melbourne: continuous and open, at the sensors only
The hourly count dataset and its companion per-minute feed are published openly, with a sensor-location file that records status and direction for each device.
Melbourne publishes a pedestrian counting system running since 2009. Sensors are installed under an awning or on a street pole and form a counting zone on the footpath below, recording bi-directional movements through that zone around the clock. The published methodology states the system records movements rather than images, so no individual information is collected. Hourly counts back to 2009 are published, alongside a per-minute feed for the most recent hour.
This is the strongest open dataset of the five - and its limit is coverage. It measures the locations where a sensor happens to be mounted, and sensors are added, moved and occasionally inactive. A site between two sensors is not measured at all.
NYC DOT: twice a year, at fixed screenlines
The bi-annual count documentation and the published count map are the clearest public statement of a sampling method we found in a large city.
New York publishes bi-annual pedestrian counts at 114 locations - 100 on-street locations, mainly retail corridors, plus bridge and greenway points. Counts are taken twice a year in May and September, on one weekday and an adjacent Saturday, in defined time bands. The documentation describes the method as screenline sampling: counts on the sidewalk mid-block, on both sides of the street.
Fifty of those locations are combined into the pedestrian volume index in the Mayor's Management Report. So a headline citywide figure rests on fifty fixed points, sampled twice a year.
Caltrans: hourly, vendor-collected, with the limits stated
The active transportation count dataset is published as CSV by year.
The California Department of Transportation publishes pedestrian and bicycle counts for state highways from district pilot projects. Data is vendor-collected using technologies including video analytics, then standardised at headquarters into CSV. The dataset is hourly and site-level, with a location identifier, timestamp, direction and count.
What makes it useful for this page is that the limitations are published rather than buried: potential undercounts in adverse weather, and incomplete coverage. Video analytics degrades in rain and at high density - the agency says so itself.
hystreet: continuous laser counting, commercial, European
Its published methodology describes the hardware and the counting line in detail.
Where a genuine 24-hour count is needed, it is being sold. One provider mounts laser scanners on building facades that generate a fourfold light curtain across the footway, measuring crossings around the clock, every day of the year, available hour by hour.
It has a free tier that exposes the current day, with hourly history behind a paid subscription - and its roughly 380 locations are in German and other European city centres. It is a real count, and it exists where the hardware is bolted to a wall.
Mapbox Movement: device density, and it says so
The movement data documentation defines the dataset and the Activity Index.
The largest footfall-sounding dataset is not a count. Mapbox Movement is built from aggregated mobile device telemetry - over 700 million devices through 45,000 apps - aggregated into 100-metre tiles, roughly one city block. Its headline metric is an Activity Index: a proprietary density value, normalised, not a number of people.
The provider's own explanation is unusually candid and worth quoting directly: it uses aggregated telemetry as a calibrated sensor to approximate people density, and states that actual ground-truth measurements of that figure simply do not exist at the spatial and temporal scales required. A density tile is a comparison signal. It is not a headcount, and treating it as one is the most common misuse in this category.
What a screenline actually is
A screenline is not an area. It is a line drawn across a footway at a chosen point, and everything counted crosses that one line. NYC DOT counts mid-block on both sides of the street at its chosen locations, which means a city figure is a sum of a fixed set of lines rather than a measurement of streets as a whole. A site can sit on a busy corridor and still be unmeasured because no line was drawn there.
Why a density tile is not a headcount
A 100-metre tile with an activity value tells you that one tile was busier than another during a time window. It does not tell you how many people entered it, and the publisher states that ground truth for that number is unavailable at that scale. The failure mode is arithmetic: comparing a density value between two tiles and calling the difference a number of visitors. It is a relative signal, and it is genuinely useful as one.
How to read any count before you trust it
Four questions separate a usable count from a number that will mislead you. Where exactly was the line? Over what period, and does that period contain the season you care about? What was the weather, and does the method degrade in it? And is the figure a count of crossings, an index normalised to a baseline, or a density estimate? A dataset that answers all four is rarer than a dataset that publishes a large file.
Why no open source gives you foot traffic for your site
Read together, the five sources explain the gap without anyone having to allege a conspiracy. Four reasons recur:
- Coverage is points, not a surface. A sensor counts one line. A camera watches one intersection. To cover a city you need hundreds of them, mounted, powered and maintained.
- Frequency hides seasonality. Counting twice a year cannot show you that a street trades differently in Ramadan, in term time, or after a competitor opens.
- Accuracy degrades exactly when it matters. Weather and density are stated limitations of video-based collection, not edge cases.
- Density is not a count. The most widely available signal is an estimate, and its own publisher says ground truth is absent at fine scales.
None of these is a defect in any particular dataset. Together they mean the number people ask for - continuous, free, city-wide, method disclosed - is not currently assembled anywhere in the open.
What open data does measure instead
Four signals are genuinely available at global coverage, each with a named dataset. None of them is a headcount, and the last column is the point of the table.
| What it measures | Dataset | What it still cannot tell you |
|---|---|---|
| Who lives in the catchment | WorldPop and ESA WorldCover | who passes the door each day |
| How far people can reach it | Valhalla and OSRM routing | who passes the door each day |
| Which competitors share the catchment | Overture Places | who passes the door each day |
| What is built there, and what the land is used for | Google Open Buildings | who passes the door each day |
How to choose a location without foot-traffic counts
Draw the catchment deliberately, then read what is inside it. Drive-time reach from real road networks is a better boundary than a distance ring, because roads, not radius, decide who arrives. Competitor proximity inside that boundary is a signal about demand that already exists. Building and land-cover data tells you whether the site is a dense retail strip or a car-dependent edge site.
And one thing cannot be substituted: actual passing behaviour. No combination of open datasets recovers how many people walk past a specific door on a Tuesday afternoon. If a decision turns on that number, it has to be counted on site, or bought from a provider with hardware in that location. Saying so is more useful than an estimate presented as a measurement.
What we do not publish
- No pedestrian counts, and no estimate of them.
- No score, ranking or recommendation is published.
- Every figure carries the dataset and version it came from.
The datasets behind our analysis are listed on the data sources page. The reasoning behind treating evidence as more valuable than a score is set out in what location intelligence means, and how the boundaries themselves are drawn is on the trade-area analysis page. Method and limits are documented under methodology.