Road checkpoint and vehicle-queue monitoring in conflict zones
Sub-metre optical imagery from commercial constellations can detect newly established road checkpoints, measure vehicle-queue length and identify barrier structures within hours of tasking. Coarser sensors provide queue-context but cannot resolve checkpoint infrastructure from ordinary roadside activity.
Sensors
- Planet SkySat: 50 cm native resolution (product sharpened to 50 cm pan, 1 m multispectral); up to 12 tasked passes per day over a given target with the full constellation of around 21 satellites. Sufficient to distinguish vehicle class (truck, SUV, armoured personnel carrier) and count individual vehicles in a queue.
- Airbus Pléiades Neo: 30 cm panchromatic, 1.2 m multispectral; stereo and tri-stereo collection in a single pass enables surface-height modelling of barrier structures and shelters. Revisit 1 to 2 days at mid-latitudes without tasking priority; same-day collection possible with priority tasking.
- Maxar WorldView-2 / WorldView-3: WorldView-3 delivers 31 cm panchromatic and 1.24 m multispectral with 8 VNIR bands plus 8 SWIR bands. SWIR is useful for camouflage discrimination. Revisit approximately 1 day at mid-latitudes. Archive depth from WorldView-2 extends to 2009, providing baseline imagery for pre-conflict road conditions.
- Planet Dove (PlanetScope): 3 to 4 m multispectral, near-daily global coverage. Cannot resolve individual checkpoint structures or vehicle type reliably. Useful for queue-length context over wider road segments, detecting traffic-flow changes across a network, and cueing sub-metre tasking to the right chokepoints.
- Copernicus Sentinel-2: 10 m multispectral, 5-day revisit at the equator (2 to 3 days at higher latitudes with both satellites). Insufficient to detect checkpoint infrastructure but valuable for mapping road-network access and identifying candidate chokepoints before sub-metre tasking is committed.
What a checkpoint looks like from 500 km up
A newly established military checkpoint leaves a recognisable signature even before any vehicle arrives. Barrier structures, whether concrete blocks, earthen berms or metal gates, cast shadows detectable at 50 cm resolution. Personnel shelters, typically canvas or prefabricated metal, appear as rectangular objects set back from the carriageway. Tyre marks diverge from the lane centre where vehicles are directed to a search bay. These features are absent in baseline imagery of the same road taken weeks or months earlier.
Vehicle queues add a second layer of evidence. At 50 cm resolution, analysts can count vehicles, distinguish broad class (heavy goods lorry, light commercial vehicle, military truck, armoured vehicle) and measure queue length in metres. A queue of 40 vehicles extending 300 m back from a point that shows no queue in prior imagery is a strong indicator of a newly imposed restriction, not a traffic accident. The distinction matters: accidents clear in minutes; checkpoints persist.
Resolution is not optional for this task
The 3 m threshold stated in the scope is not arbitrary. At 3 m per pixel, a standard passenger car occupies roughly two pixels in each dimension. Vehicle counting becomes unreliable, vehicle-class discrimination is near-impossible, and small barrier structures merge with road markings or roadside clutter. Published comparisons of vehicle-detection performance across resolution classes consistently show that reliable individual-vehicle detection requires pixels no larger than approximately 1 m, with sub-50 cm imagery needed for class discrimination in mixed civilian-military queues.
This has a practical consequence for collection strategy. Planet Dove or Sentinel-2 can flag that something has changed at a road junction, but they cannot confirm what. Sub-metre tasking must follow. The latency between the Dove flag and the SkySat or Pléiades Neo collect is the operationally critical interval, and it is typically measured in hours, not days, when tasking priority is pre-arranged.
Change detection as the analytic backbone
The core method is object-level change detection against a pre-event baseline. Analysts, or automated pipelines trained on labelled imagery, compare current sub-metre scenes against the most recent clean baseline of the same road segment. Objects appearing in the current scene that are absent from the baseline, and that match the spectral and geometric signatures of barriers, shelters or queued vehicles, are flagged as candidate checkpoint features.
Automated vehicle detection using convolutional neural networks trained on satellite imagery has been published extensively in the remote-sensing literature. Detection rates for vehicles in open areas at 50 cm resolution exceed 90 percent under good conditions in peer-reviewed studies, though performance degrades under partial shadow, camouflage netting or when vehicles are closely packed. Human analyst review remains necessary for ambiguous cases, particularly where civilian and military vehicles are intermingled.
Temporal stacking adds confidence. A single collect showing a queue could reflect a road accident or a market day. Three collects over 48 hours showing a persistent queue at the same point, with the same barrier geometry, constitutes a much stronger indicator of a deliberate restriction. High-revisit constellations make this kind of temporal triangulation operationally feasible.
Honest limits: what the imagery cannot tell you
Cloud cover is the obvious constraint. Optical sensors see nothing through cloud, and conflict zones in tropical or monsoon climates can be obscured for days at a time. SAR can detect large metallic objects and surface disturbance through cloud, but at the resolutions currently available from Sentinel-1 (5 by 20 m in IW mode) or commercial SAR providers, individual vehicle detection is not reliable. SAR is a complement, not a substitute, for optical at this task.
Imagery confirms presence and geometry but not intent. A queue of vehicles at a road barrier could indicate a military checkpoint, a customs post, a humanitarian aid distribution point or a civilian evacuation control measure. Context from other intelligence sources is required to assign meaning. Analysts should report what the imagery shows, not what it implies without corroboration.
Night-time optical collection is possible with SkySat and Pléiades Neo under moonlit conditions, but sensitivity is reduced and vehicle-class discrimination becomes harder. Thermal infrared from sensors such as Landsat 8/9 TIRS or ECOSTRESS can detect engine heat from idling vehicles at night, though at 100 m resolution for TIRS, individual vehicles are not resolved. Thermal is useful for confirming that a queue is active rather than abandoned.
From raw collect to decision-ready product
A practical monitoring workflow runs in three tiers. The first tier uses near-daily Dove or Sentinel-2 coverage to watch a defined road network for anomalous traffic signatures, triggering a sub-metre task order when a candidate location is identified. The second tier delivers the sub-metre collect, orthorectified and pansharpened, within a few hours of acquisition. The third tier applies change detection against the baseline and produces a structured report: checkpoint confirmed or not confirmed, vehicle count, queue length in metres, barrier type, and a georeferenced polygon of the affected road segment.
Satellize runs analytics of this type on open and commercial imagery, with the sub-metre commercial layer added on client licence. The output is a GIS-compatible layer and a structured alert feed rather than a raw image dump, which is what operational users actually need at tempo.
Archive exploitation adds a dimension that real-time monitoring cannot. Maxar's WorldView-2 archive extends to 2009. Pléiades archive coverage of many conflict-affected regions goes back to 2012. Systematic review of archive imagery can reconstruct when a checkpoint first appeared, how its infrastructure evolved and whether it has been used before, all of which are relevant to legal and accountability analysis as well as operational assessment.
Typical figures
| Best available spatial resolution | 30 cm panchromatic (Airbus Pléiades Neo); 31 cm panchromatic (Maxar WorldView-3); 50 cm (Planet SkySat) |
| Minimum detectable target | Individual vehicle reliably detected at ≤1 m resolution; vehicle class discrimination requires ≤50 cm |
| Revisit rate (sub-metre, tasked) | Up to 12 passes per day (SkySat full constellation); 1 to 2 days standard, same-day priority (Pléiades Neo); ~1 day (WorldView-3) |
| Revisit rate (context, untasked) | Near-daily (PlanetScope Dove, 3 to 4 m); 2 to 5 days (Sentinel-2, 10 m) |
| Spectral bands | Pan + 4-band VNIR standard; 8-band VNIR + 8-band SWIR available (WorldView-3) for camouflage discrimination |
| Latency from collect to delivery | Typically 2 to 6 hours for priority-tasked commercial imagery; same-day for open Sentinel-2 |
| Cloud cover constraint | Optical sensors fully blocked by cloud; no reliable workaround at vehicle-detection resolution from current SAR |
| Archive depth | WorldView-2 from 2009; Pléiades from 2012; Sentinel-2 from 2015; PlanetScope from 2016 |
| Coverage per collect | SkySat strip up to 20 km × variable; Pléiades Neo swath 14 km; WorldView-3 swath 13.1 km |
| Delivery formats | Orthorectified GeoTIFF; GeoJSON or Shapefile change-detection layer; structured alert feed (JSON); PDF report |
Analytics Satellize can run
| Checkpoint appearance alert | Object-level change detection against pre-event baseline; threshold on new objects matching barrier and shelter spectral-geometric signatures | Georeferenced alert with confidence score, timestamp, and annotated image chip; delivered as JSON feed or email notification |
| Vehicle count and queue-length measurement | CNN-based vehicle detection trained on sub-metre satellite imagery; queue length computed as distance from leading vehicle to barrier object | Structured table: vehicle count, queue length in metres, vehicle-class breakdown, collect time; updated per new collect |
| Checkpoint persistence and temporal profile | Multi-date stack analysis across all available collects; presence/absence scored per collect to build a time-series | Time-series chart and GIS layer showing checkpoint active/inactive status by date; exportable to CSV |
| Barrier and shelter infrastructure classification | Object segmentation and shape analysis at 30 to 50 cm resolution; comparison against library of known barrier types | Annotated orthophoto with labelled infrastructure objects and dimensional estimates; PDF or GeoPackage |
| Road-network chokepoint prioritisation | Network analysis on road-graph combined with Dove or Sentinel-2 traffic-anomaly flags to rank candidate locations for sub-metre tasking | Ranked tasking list updated daily; GIS layer of candidate chokepoints with anomaly score |
| Archive-based checkpoint history reconstruction | Systematic review of WorldView-2/Pléiades archive imagery at candidate locations; manual and semi-automated change detection across multi-year stack | Chronological report of checkpoint first appearance, infrastructure evolution and prior use periods; suitable for accountability documentation |
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.