Pavement and carriageway encroachment detection by mobile structures
Multi-date very-high-resolution imagery from SkySat and Pleiades Neo can detect semi-permanent kiosks, canopies and stalls occupying pavements or traffic lanes, giving enforcement teams a spatially precise, time-stamped record of encroachment.
Sensors
- Planet SkySat: 0.5 m panchromatic, 0.72 m multispectral (four bands). Daily revisit over tasked areas makes it practical to build a dense time series and catch structures that appear and disappear within 24 to 48 hours.
- Airbus Pleiades Neo: 0.3 m panchromatic, 0.75 m multispectral (six bands including red-edge). The additional spectral depth helps distinguish tarpaulin and painted metal from road markings that share similar geometry.
- Maxar WorldView-3: 0.31 m panchromatic, 1.24 m multispectral (eight bands) plus eight SWIR bands. SWIR is particularly useful for separating plastic sheeting from concrete by reflectance signature, though revisit is lower than SkySat.
- Maxar WorldView Legion: 0.29 m panchromatic with planned revisit of up to six times per day over priority cities. Still being commissioned as of 2024, but the revisit cadence, once stable, would suit rapid enforcement cycles.
What a tarpaulin looks like from 500 km up
Paved road surfaces are spectrally dull. Asphalt sits in a narrow reflectance band across visible wavelengths, and concrete is only marginally brighter. Vendor structures are not dull. Blue polyethylene tarpaulins, orange woven sacking, galvanised corrugated iron and painted timber all have reflectance signatures that stand out sharply against carriageway material, particularly in the red and near-infrared bands available on Pleiades Neo and WorldView-3.
Geometry reinforces spectral contrast. A 3 m by 4 m market stall produces a rectangular shadow at oblique sun angles and a flat, high-reflectance roof patch at nadir. At 0.3 to 0.5 m ground sample distance, individual stalls are resolvable as discrete objects. A cluster of ten stalls occupying a bus lane is detectable even before any change-detection algorithm is applied; a single stall at the edge of a pavement requires careful baseline subtraction to separate from legitimate street furniture.
Building a clean-road baseline and measuring what changes
The analytic method rests on a reference image acquired when the road is clear, typically early morning on a non-market day, or from a historical archive date confirmed by ground inspection. Subsequent acquisitions are co-registered to the baseline at sub-pixel accuracy, a step that matters enormously at 0.3 to 0.5 m resolution where a one-pixel shift would fabricate or erase a stall-sized object.
Change detection then operates on the difference image. Spectral change indices, including simple band-ratio differencing and more structured approaches such as iteratively reweighted multivariate alteration detection (IR-MAD, described in peer-reviewed remote sensing literature), flag pixels whose reflectance has shifted beyond a calibrated threshold. The flagged patches are then filtered by shape: elongated or irregular blobs are more likely to be shadow artefacts or vegetation; compact rectangles matching the expected footprint of market structures are retained as candidate encroachments.
The practical detection floor sits at roughly 4 to 6 square metres at 0.5 m GSD, assuming good radiometric calibration and a cloud-free acquisition. Smaller temporary objects, a single umbrella or a bicycle cart, fall below reliable detection. Clients should treat the method as a systematic screen for semi-permanent structures, not a census of every individual vendor.
Temporal density is the operational argument
A single snapshot tells an enforcement officer that a structure exists today. A time series tells them when it appeared, how often it is present, and whether it has grown. SkySat's daily tasking cadence over a defined area of interest makes that time series achievable at reasonable cost. Pleiades Neo can be tasked for same-day or next-day delivery in most urban markets.
In practice, a weekly acquisition rhythm is sufficient for most enforcement programmes: enough temporal resolution to distinguish genuinely temporary market-day activity from structures that are effectively permanent. Daily tasking adds value when a municipality needs to document specific dates of violation for legal proceedings, or when a road-widening project requires a before-and-after record at high temporal precision.
Cloud cover is the honest constraint. In humid tropical cities, consecutive cloudy days can break a time series for days at a time. Tasking multiple sensors simultaneously, SkySat and Pleiades Neo over the same corridor, reduces the probability of a complete gap but does not eliminate it. Radar alternatives such as Sentinel-1 SAR are cloud-immune but cannot resolve individual stalls at their 5 to 20 m resolution; they are useful for tracking broader carriageway obstruction patterns, not individual structures.
From pixel flags to enforcement-ready outputs
Raw change masks are not enforcement documents. The analytic pipeline converts pixel-level detections into georeferenced polygons, each attributed with a first-detection date, an estimated footprint area in square metres, a confidence score, and, where the time series is dense enough, a persistence score indicating how many acquisition dates the structure was present.
These polygons are delivered as GIS layers in standard formats (GeoJSON, shapefile, or GeoPackage) that can be loaded directly into a municipality's existing spatial planning or enforcement management system. A tabular summary report ranks encroachment hotspots by total area occupied and by persistence, giving enforcement teams a prioritised work list rather than an undifferentiated map of thousands of flagged pixels.
Integration with street-address databases or cadastral layers allows each detected structure to be associated with a specific plot frontage or road segment, which is the unit that most enforcement regulations reference. That linkage is done in post-processing and depends on the quality of the client's own cadastral data.
Honest limits and what they mean for programme design
Several failure modes deserve direct acknowledgement. First, structures with roofing materials that spectrally resemble paving, weathered grey tarpaulins, bare concrete slabs used as market tables, will generate false negatives. Spectral indices alone cannot catch everything; periodic ground-truth sampling is necessary to calibrate detection confidence for a given city's vendor population.
Second, the method cannot distinguish a licensed kiosk from an unlicensed one. Satellite imagery records physical presence, not legal status. The enforcement value comes from identifying what is there and when it appeared; the legal determination still requires cross-referencing with permit databases held by the client.
Third, very dense informal market areas, where structures are packed wall to wall, reduce the geometric distinctiveness that makes individual detection reliable. The method works best on arterial roads and designated pavements where the baseline surface is clearly paved and structures are discrete. In a market square that is entirely covered by canopies, the baseline itself may be ambiguous.
Satellize has applied similar spectral-change and object-detection pipelines in its analytics work, including the Kingdom of Tonga crop-estimation programme, and the same co-registration and change-flagging logic transfers directly to urban encroachment contexts. The core method is not exotic; its value lies in systematic, repeatable application across a city at a cadence that manual inspection cannot match.
Typical figures
| Spatial resolution (panchromatic) | 0.29 m (WorldView Legion), 0.3 m (Pleiades Neo), 0.31 m (WorldView-3), 0.5 m (SkySat) |
| Spatial resolution (multispectral) | 0.72 m (SkySat), 0.75 m (Pleiades Neo), 1.24 m (WorldView-3) |
| Revisit cadence | Daily (SkySat over tasked area); 1 to 2 days (Pleiades Neo); up to 6× daily (WorldView Legion, commissioning ongoing) |
| Spectral bands relevant to this use case | Blue, Green, Red, NIR (all sensors); Red-edge (Pleiades Neo); SWIR bands 1–8 (WorldView-3) |
| Minimum detectable encroachment footprint | Approximately 4 to 6 m² at 0.5 m GSD under good radiometric conditions; smaller structures unreliable |
| Delivery latency | 24 to 72 hours from acquisition for processed change-detection layer; same-day feasible for priority tasking |
| Archive depth | SkySat archive from 2016; Pleiades (original) from 2012; WorldView-3 from 2014 |
| Cloud constraint | Optical sensors only; cloud cover above 20 to 30 % typically voids an acquisition; multi-sensor tasking reduces gap probability |
| Delivery formats | GeoJSON, Shapefile, GeoPackage (vector detections); GeoTIFF (change rasters); PDF tabular summary report |
| Coverage per tasking pass | SkySat strip: approximately 6.6 km wide; Pleiades Neo scene: 14 km × 14 km; can be mosaicked for city-wide coverage |
Analytics Satellize can run
| Clean-road baseline map | Manual and automated selection of cloud-free, low-occupancy acquisition; radiometric normalisation to surface reflectance | GeoTIFF baseline mosaic with acquisition date metadata, used as reference for all subsequent change runs |
| Bi-weekly encroachment change mask | Co-registered band-ratio differencing and IR-MAD change detection against baseline; morphological filtering to remove shadow and vegetation artefacts | GeoJSON polygon layer of flagged encroachments, attributed with detection date, area (m²) and confidence score |
| Persistence and recurrence scoring | Temporal stack analysis counting acquisition dates on which each flagged polygon is present; normalised to a 0 to 1 persistence index | Attributed GIS layer ranking structures by permanence; CSV summary table for enforcement prioritisation |
| Encroachment hotspot report | Spatial aggregation of flagged polygons by road segment or administrative zone; ranking by total occupied area and persistence | PDF and spreadsheet report listing top encroachment corridors with coordinates, estimated footprint and first-detection date |
| Before-and-after evidence pack | Paired image chips (baseline vs. detection date) clipped to each flagged polygon, with co-registered overlay | Zipped folder of georeferenced image pairs suitable for inclusion in enforcement or legal documentation |
| Road-segment occupancy time series | Longitudinal tracking of total encroached area per named road segment across all available acquisition dates | Interactive chart (HTML or PDF) and underlying CSV showing area occupied over time, exportable for traffic management reporting |
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.