Outdoor advertising and billboard inventory mapping

Sub-metre optical satellites and oblique aerial imagery can detect, locate, and height-estimate large outdoor advertising structures across an entire city in days, exposing the gap between what operators have licensed and what is actually standing.

Sensors

WorldView-3: Panchromatic resolution of 0.31 m and eight-band SWIR, enabling both object detection and spectral contrast analysis of retroreflective or painted sign faces. Revisit roughly 1 day at mid-latitudes with off-nadir tasking.
Pleiades Neo: 0.30 m native panchromatic resolution, four-band plus deep-blue multispectral, with stereo and tri-stereo collection modes that support direct 3-D structure modelling of billboard frames without relying solely on shadow geometry.
Oblique aerial imagery (e.g. Nearmap, Cyclomedia, Vexcel): Ground sampling distances of 5 to 15 cm from aircraft at typical survey altitudes. Captures sign faces, structural frames, and mounting hardware in perspective view, resolving detail that nadir satellite imagery misses entirely in street canyons.
SkySat: 0.50 m panchromatic, with a rapid-revisit constellation capable of collecting the same city block multiple times per day. Useful for change detection between audit cycles rather than initial high-precision inventory.

What a shadow gives away

A billboard standing on a pole or parapet casts a shadow whose length is a direct function of solar elevation angle, the structure's height, and the local terrain. With the sun's position known to arc-second precision for any acquisition time, and with the ground plane extracted from a city digital terrain model, shadow length becomes a height estimator. On WorldView-3 imagery at 0.31 m, shadows from structures taller than roughly two metres are reliably measurable. The practical floor for height estimation by this method is around three metres, below which shadow measurement error from pixel quantisation becomes too large to be useful.

The geometry is straightforward: H = L × tan(θ), where H is object height, L is shadow length measured in image space and converted to ground distance, and θ is the solar elevation angle at acquisition. Errors compound from three sources: the accuracy of the ground plane model, the precision with which the shadow tip can be identified in the image, and any slope in the underlying surface. On flat urban hardstanding, total height uncertainty is typically plus or minus 0.5 to 1.5 metres for structures in the five-to-fifteen-metre range. That is sufficient to distinguish a standard six-sheet panel from a large-format 48-sheet hoarding, which matters for fee schedules that are banded by face size and height.

Spectral contrast and object detection: finding the sign face itself

Shadow geometry locates the base of a structure. Detecting the sign face directly requires a different approach. Advertising panels are among the most spectrally anomalous objects in the urban scene. Vinyl substrates, retroreflective sheeting, and illuminated LED arrays all produce reflectance signatures that differ sharply from roofing materials, road surfaces, and vegetation. In multispectral imagery, the ratio of near-infrared to red reflectance for a white or light-coloured advertising face is distinctively high compared with surrounding concrete or asphalt. This spectral contrast is exploitable as a first-pass filter before object detection models are applied.

Convolutional neural network detectors trained on labelled sub-metre imagery can identify billboard structures with precision and recall figures in the high eighties to low nineties on well-lit, unoccluded examples, based on published work in the remote sensing literature. Performance degrades on heavily shaded north-facing faces, on structures partially occluded by tree canopy, and on small-format signs whose physical footprint approaches the resolution limit. A 0.31 m pixel means a sign face of one metre width spans roughly three pixels; detection at that scale is unreliable. The practical minimum detectable sign face for automated detection on WorldView-3 or Pleiades Neo is approximately two metres in the smaller dimension.

Street canyons: the hard limit of nadir imagery

In dense urban cores, nadir satellite imagery fails for a straightforward geometric reason. A billboard mounted on the wall of a building at two-thirds of its height, facing along a narrow street, may be entirely occluded by the building's own parapet when viewed from directly above. The structure exists; the sensor cannot see it. This is not a resolution problem. It is a view-angle problem, and no increase in satellite resolution resolves it.

Oblique aerial imagery addresses this directly. Surveys flown at four or five look directions, combined with a nadir pass, produce a near-complete perspective coverage of vertical building faces. Providers such as Nearmap and Vexcel Imaging supply city-wide oblique datasets at ground sampling distances of five to fifteen centimetres, sufficient to read sign text and identify mounting hardware. For a municipal licensing audit, the honest workflow combines satellite imagery for rapid city-wide triage with oblique aerial data for street-canyon reconciliation. Relying on satellites alone will undercount, and the undercount will be systematically biased towards dense central districts where unregistered signage is often most concentrated.

Building the inventory: from detections to a licence audit

Raw detections from imagery are points on a map. A licence audit requires those points to be matched against the authority's existing register, geocoded to cadastral parcels, attributed with estimated dimensions and height, and flagged as either registered, unregistered, or ambiguous. The matching step is where most of the analytical effort sits. Positional accuracy of detections from WorldView-3 with rational polynomial coefficient correction is typically one to three metres CE90, which is sufficient to assign a detection to a parcel but not always to a specific building face in a dense block.

Stereo collection from Pleiades Neo can tighten this considerably. A stereo pair yields a digital surface model accurate to roughly 0.5 metres vertically in open areas, allowing the top of a billboard structure to be located in three dimensions and matched to a specific parapet or pole. For authorities that want a repeatable audit cycle, the most cost-effective architecture is a baseline oblique aerial survey for the initial inventory, then quarterly or annual satellite tasking for change detection, flagging new structures or removals for field verification. The satellite pass does not replace the oblique survey; it extends its useful life.

Revenue recovery: what the numbers actually look like

The commercial case for satellite-assisted billboard audits rests on a simple observation: in most cities with active outdoor advertising markets, the gap between registered and physically present structures is significant. Studies and audits in various jurisdictions have found unregistered rates ranging from fifteen to over forty per cent of total physical inventory, though these figures vary widely by city and enforcement history. Satellize does not publish a standard figure because the range is too wide to be honest as a single claim.

What satellite and aerial imagery can do is produce a defensible, dated, spatially referenced inventory that an authority can use as evidence in enforcement proceedings. A detection accompanied by a georeferenced image chip, a shadow-derived height estimate, and a parcel attribution is a stronger basis for a licence demand than a field officer's notebook entry. The Satellize analytics workflow for this use case delivers a GIS layer with those attributes attached, structured for direct import into a municipal asset management or licensing system. The Kingdom of Tonga crop-estimation programme demonstrated a comparable principle: that a small analytics team working on commercial satellite data can produce audit-quality outputs that a government can act on without building its own satellite infrastructure.

Honest limits, plainly stated

Automated detection on sub-metre imagery will miss structures in deep shadow, beneath tree canopy, and in narrow street canyons. It will generate false positives on rooftop plant, ventilation housings, and certain solar panel arrays whose spectral and geometric signatures overlap with advertising structures. A post-processing review step, either human or assisted by a secondary classifier, is not optional. Expect to spend roughly as much time on quality control as on the initial detection run.

Cloud cover is a constraint for optical sensors. A city with persistent seasonal overcast may require multiple tasking attempts over weeks to achieve adequate coverage. Oblique aerial surveys are similarly weather-dependent. Synthetic aperture radar can detect large metallic structures through cloud, but the interpretation of SAR backscatter for thin vertical objects like billboard frames is ambiguous and not yet operationally mature for this application. For now, optical imagery is the operative method, and cloud is a scheduling problem rather than a technical one.

Typical figures

Best available spatial resolution (satellite)	0.30 m panchromatic (Pleiades Neo, WorldView-3)
Best available spatial resolution (oblique aerial)	5 to 15 cm GSD depending on aircraft altitude and sensor
Satellite revisit (tasked)	1 day or less at mid-latitudes for WorldView-3 and Pleiades Neo with off-nadir collection
Positional accuracy (satellite, with RPC correction)	1 to 3 m CE90 for WorldView-3 and Pleiades Neo without ground control points
Minimum detectable sign face (automated)	Approximately 2 m in smallest dimension on 0.30 to 0.31 m imagery
Height estimation uncertainty (shadow method)	±0.5 to 1.5 m for structures 5 to 15 m tall on flat terrain
Spectral bands used	Panchromatic, red, green, blue, near-infrared; SWIR (WorldView-3) for material discrimination
Archive depth	WorldView-3 from 2014; Pleiades Neo from 2021; historical oblique surveys vary by provider and city
Delivery format	GeoJSON or shapefile point/polygon layer with attributes; image chips per detection; height and dimension estimates

Analytics Satellize can run

City-wide billboard detection layer	Convolutional neural network object detection on pan-sharpened sub-metre optical imagery, with spectral contrast pre-filter on NIR/red ratio	GeoJSON point layer, one feature per detected structure, with confidence score and image chip
Height and dimension estimates per structure	Shadow length measurement in image space combined with solar ephemeris and DTM-derived ground plane; stereo DSM for Pleiades Neo stereo pairs	Attribute table appended to detection layer: estimated height, shadow length, solar elevation at acquisition, estimated face width where resolvable
Licence register gap analysis	Spatial join of detected structures to municipal licence register polygons or points; unmatched detections flagged as potentially unregistered	Tabular report and GIS layer with three-class attribution: registered, unregistered, ambiguous; summary counts by administrative zone
Change detection between audit cycles	Image differencing and re-detection on repeat satellite tasking; new detections and disappearances flagged against baseline inventory	Change layer with date-stamped additions and removals; alert feed configurable by zone or parcel
Street-canyon reconciliation layer	Oblique aerial image analysis for building-face mounted signage not visible in nadir imagery; manual or semi-automated digitisation of sign faces from multi-directional oblique views	Supplementary GeoJSON layer for street-level structures; merged with satellite detection layer for complete city inventory
Revenue exposure estimate by zone	Detected unregistered structures attributed with estimated face area and height band; multiplied against authority fee schedule supplied by client	Spreadsheet report: estimated annual licence revenue recoverable per zone, with uncertainty range based on detection confidence and dimension estimation error

Sources

Who does the work

We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.