Safest Way is a pedestrian navigation app that does something most map apps don't bother with: it treats a dark alley and a lit high street as genuinely different places, even when they're the same distance from A to B. The app routes people towards streets that are actually lit and watched, not just streets that are short.

Building reliable, safety-conscious navigation relies entirely on granular data detailing the exact placement of streetlights and CCTV cameras. That sounds like a solved problem; surely a local authority has a spreadsheet somewhere? For most places, it is. But for a worrying number of places, it isn't.

Getting Data

Our first move for any new city is always the same: file a Freedom Of Information (FOI) request and ask the council directly. We've done this for borough after borough across London and major UK cities. When it works, it's the best data we'll ever get, councils maintain asset registers for street lighting because they pay the electricity bill and send crews out when a lamp dies. That obligation produces complete coverage, accurate coordinates, and a maintenance cycle that keeps the data fresh.

Around 113,000 of the 1.9 million* lighting sources in our UK database come from Mapillary, filling gaps that council records don't cover.

*data as of June 2026

CCTV is a different story. Some councils hand over camera locations without fuss, Edinburgh even publishes them in interactive maps. Others have static PDFs from GIS software. But some won't share at all, citing security grounds: a public map of every camera is also a map of where the gaps are. Even though we promise never to share actual locations, only to use them for safer routing, arguing the point usually gets us nowhere.

There's a messier problem underneath: many cameras on any given street aren't council assets at all. They belong to shops, residents' associations, business improvement districts, and private security firms, no single register exists. Then there are streets where councils aren't in charge of the infrastructure at all; large stretches of London's South Bank and King's Cross, for instance, are maintained by business improvement districts. For these areas, no FOI route, no public owner to ask, we needed a way to go and look for ourselves.

Going and looking for ourselves

This is the part where Mapillary stopped being a "nice to have" and became infrastructure. London, where Safest Way started, is one of the better mapped capital cities on the platform, so for many central streets and peripheral main roads there is street view imagery available, contributed by people who've been driving or walking around with a camera mounted on a dashboard or a helmet, often for reasons that have nothing to do with street safety.

Street-level imagery coverage across Central London on Mapillary.

We don't use street view imagery directly for this; we use the Mapillary API instead. (Street imagery itself we do use heavily, for safety perception, that's a separate pipeline.) Mapillary processes the imagery using computer vision to build 3D scenes. Within roughly a day of upload, it automatically identifies map features, including street lights and CCTV cameras, as point locations queryable through the API. We simply send a request for a given area and get back a list of detections, each complete with its feature type and geographic coordinates.

Where we want to accelerate coverage, we capture our own footage with a GoPro MAX mounted as Mapillary recommends. At cycling speed, an hour covers around 10 km and yields thousands of detected features once processed, a fast way to extend coverage rather than waiting for community imagery to catch up.

Between community coverage and the gaps we fill ourselves, this replaces what used to be manual point-collection (someone walking a street, and logging coordinates by hand) with:

1 - Find or capture the imagery,
2 - Wait a day for processing,
3 - Call the API for the detections

Computer vision supported the workflow by automating the feature extraction, saving us hours in data processing and mapping.

What sold us is speed at scale. Once imagery has been processed, we can query an entire city's worth of detected street lights and CCTV cameras in minutes. Everything downstream on our side, filtering, deduplicating, fusing with other layers, runs in under an hour. For a small team covering multiple cities, that turnaround is the difference between "now" and "maybe next year".

Safest Way's London Lighting Map, built with Mapillary and council data. Explore it at safestway.co.uk/london-lighting-map.

Making the data work for us

Two things are worth knowing if you want to get the most out of the API's detailed output.

First, GPS drift. A contributor's track isn't always precisely geolocated, so detected objects can land a few metres from their true position. We mitigate this by favouring tracks that align with known road geometry, weighting newer footage over older, and cross-checking against council data where available. No single check fully resolves drift, but together they push detections closer to reality before the thinning step below.

Second, and more relevant for us: popular streets accumulate overlapping tracks over time, so the same lamp post or camera can appear as multiple separate points from different passes. We treat this as a feature, not a bug. The API returns raw detections without deduplicating them, which means we can apply our own spatial thinning logic tuned to what we know about street furniture, rather than relying on rules we can't see or control. We've built a deduplication step that strips out near-neighbour detections that are almost certainly the same physical object.

One layer among several

Mapillary is one of many data sources we fuse through the Safest Way API, sitting alongside council-supplied lighting and CCTV records, crime and public infrastructure data. What Mapillary gives us specifically is coverage in the places official records don't reach, such as privately managed estates where the asset owner isn't a public body with an FOI obligation.

For a small team building safety infrastructure across multiple UK cities, that's a meaningful gap to have closed.

From imagery to safe routing: the Safest Way data pipeline

Safest Way is a pedestrian safety navigation platform supported by Ordnance Survey's Geovation and the UCL Impact accelerator. The mobile app, available for free on iOS and Android, is live in London, Northern Ireland, Birmingham, York and other parts of the UK. We're actively expanding our coverage, and last month we passed 15,000 mobile app installations.

The product improves safety in two linked ways. First, the routing itself: streets are scored using street lighting and CCTV coverage, official crime data, and urban features such as parks and derelict buildings, so the app routes people towards corridors that are brighter and more likely to be watched or busy, not just shorter. Second, perceived safety, which doesn't always track objective data one-to-one: a well-lit street can still feel unsafe depending on how it looks and how it's used. We model this using vision-language techniques trained on how a street actually reads to someone walking it, rather than relying purely on infrastructure counts.

Mapillary sits at the foundation of that first layer, giving us the asset data we need in places no official record covers. For a platform built on the idea that where you walk matters as much as how far, that's not a nice extra, its safety.

Learn more about Safest Way and our API on safestway.co.uk.

Ilya Ilyankou

Alesja Gilvear