Video from space for transient activity detection
Commercial video satellites capture short clips at sub-metre resolution, revealing moving vehicles, vessel wakes and aircraft invisible in still imagery. Frame rate, footprint and data volume impose hard limits that any serious programme must plan around.
Sensors
- Planet SkySat: Captures video clips of roughly 30 seconds at approximately 0.9 m ground sample distance, 30 frames per second, in a footprint of about 2 km × 1.1 km. The constellation of around 21 satellites enables tasking at high latitudes multiple times per day, though any single clip covers a small area.
- Satellogic VideoSat: Produces video at approximately 1 m resolution and up to 30 fps. Satellogic has published clips demonstrating moving vehicle detection; footprint per clip is similarly constrained to a few square kilometres.
- BlackSky: BlackSky's Gen-3 satellites offer sub-metre panchromatic imaging and have demonstrated short video burst capability. The constellation is designed for rapid revisit of specific sites, with tasking response times that can fall below 90 minutes for priority requests.
- Planet SkySat still vs. video mode: SkySat can operate in still-frame collect mode at 50 cm (with pansharpening) or video mode at 0.9 m. Operators choose: video costs more data volume and tasking time, still imagery gives sharper spatial detail. The trade-off is explicit and consequential.
What a moving object leaves behind that a still image cannot show
A single optical frame is a frozen moment. A vehicle parked beside a warehouse and a vehicle actively loading cargo look identical in it. Video changes that. At 30 frames per second and 0.9 m ground sample distance, a truck travelling at 30 km/h moves roughly 0.25 m between frames, which is well below one pixel. Detection therefore does not rely on tracking pixel displacement frame to frame; it relies on the persistence of the object across many frames and on the wake or disturbance it leaves in the scene.
Vessel wakes are the clearest example. A ship underway at 15 knots generates a Kelvin wake extending hundreds of metres behind it. Even in a single frame that wake is visible, but video confirms its direction of travel, estimates speed from wake angle, and distinguishes a vessel that stopped from one still moving. Aircraft on approach leave a similar signature: a consistent bright point moving at predictable velocity across a runway threshold. These are not exotic detections. They are routine outputs of frame-differencing algorithms applied to clips that any competent analyst can run.
Frame rate and compression: the physics of what you can and cannot resolve
Thirty frames per second sounds fast. Against the velocities that matter in space surveillance it is not always enough. A commercial aircraft on approach at 250 km/h moves roughly 2.3 m per frame at 0.9 m GSD, which is about 2.5 pixels. That is detectable. A fast-moving speedboat at 60 km/h moves about 0.56 m per frame, close to the resolution floor. At that point, detection becomes probabilistic rather than certain, and it degrades further when compression artefacts are introduced.
All commercial video satellites apply lossy compression before downlink. The raw data rate for uncompressed 30 fps video at 0.9 m GSD over a 2 km × 1.1 km footprint is substantial, far exceeding what most ground-station passes can absorb. Compression ratios vary by operator and are not always published, but artefacts around high-contrast edges (a white wake on dark water, a bright aircraft against tarmac) are a known and documented limitation. Analysts working with space video should validate compression settings for their specific tasking before committing to a detection threshold.
The footprint problem: small clips, large world
A SkySat video clip covers roughly 2.2 km². Greater London is approximately 1,572 km². Covering it with video in a single pass is not possible. This is the central operational constraint of space video that marketing materials tend to understate.
Effective use of space video therefore requires prior intelligence. You task video against a specific berth, a known airstrip, a border crossing, a suspected storage yard. The clip confirms or denies activity at that location. It is not a surveillance net cast wide; it is a pointed question asked of a known address. Programmes that treat it otherwise burn tasking budget rapidly and return inconclusive data. The discipline of defining the question before ordering the clip is not optional.
Data volume and the cost of routine tasking
A single 30-second SkySat video clip at full quality runs to several gigabytes before any processing. Order ten clips per day across five sites and the ingest, storage and processing pipeline becomes a genuine engineering problem, not an afterthought. Cloud processing mitigates storage costs but introduces latency if the analytic chain is not pre-built and automated.
Revisit frequency compounds this. BlackSky advertises sub-90-minute tasking response for priority customers. Planet's SkySat constellation can revisit a mid-latitude site multiple times per day. If each visit generates a clip, the data volume scales accordingly. Programmes that want near-real-time activity monitoring at multiple sites simultaneously need automated frame-differencing pipelines, not human analysts watching footage. The human role shifts to exception review: confirming what the algorithm flagged, not watching everything.
Where video beats SAR, and where it does not
Space video is optical. Cloud cover blocks it completely. A persistent cloud deck over a port of interest can deny collection for days. Synthetic aperture radar, covered on a sibling page, sees through cloud and operates day or night, but it does not produce intuitive visual clips and its moving-target indication capability is more complex to interpret. The two modalities are complements, not substitutes.
Video also requires daylight. Twilight collections are possible but image quality degrades. Night video from current commercial systems is not operationally useful at sub-metre resolution; that gap is partly addressed by VIIRS day-night band monitoring, which is covered separately and operates at far coarser resolution. For a programme that needs continuous, all-weather activity monitoring, space video is one layer in a stack, not the whole answer.
Building an analytic programme around space video
The practical architecture for a government or enterprise customer starts with a site list, ranked by priority. Each site gets a tasking cadence matched to the operational question: daily for an active port, weekly for a construction site, on-demand for an event-driven trigger. Automated frame-differencing or optical-flow algorithms run on ingest and generate alerts for human review. Analysts see flagged frames, not raw footage.
Satellize integrates commercial video tasking into client programmes on a licensed basis, building the ingest and alerting pipeline so that the client receives structured outputs rather than raw clips. The analytic methods rest on published optical-flow and background-subtraction techniques well documented in the remote-sensing literature. For clients considering this capability, the right starting conversation is about the site list and the detection threshold, not the satellite. Book a technical scoping call to work through those two questions.
Typical figures
| Spatial resolution (video mode) | 0.9 m GSD (Planet SkySat); approximately 1 m (Satellogic); sub-metre panchromatic (BlackSky Gen-3) |
| Frame rate | Up to 30 fps (SkySat, Satellogic); actual delivered rate depends on compression and downlink conditions |
| Clip duration | Typically 30 to 90 seconds per tasked collect; longer clips increase data volume and reduce revisit opportunity |
| Footprint per clip | Approximately 2 km × 1.1 km (SkySat); varies by satellite altitude and sensor field of view |
| Spectral bands | Panchromatic (primary for video); some systems add RGB colour at reduced frame rate |
| Minimum detectable moving target | Vehicles above roughly 5 m length travelling faster than approximately 10 km/h; smaller or slower objects approach the resolution and compression floor |
| Revisit / tasking latency | Multiple times per day at mid-latitudes (SkySat constellation); BlackSky priority tasking response under 90 minutes published |
| Cloud limitation | Complete optical blockage; no cloud-penetration capability |
| Data volume per clip | Several gigabytes uncompressed; compressed delivery varies by operator and is not always published |
| Archive depth | Planet SkySat video archive available from approximately 2015; depth and completeness vary by location |
Analytics Satellize can run
| Moving vehicle count and direction | Optical flow and frame-differencing applied to sequential video frames; background subtraction to isolate moving foreground objects | Structured count report per clip, with direction vectors and estimated speed ranges, delivered as JSON or CSV |
| Vessel underway / at anchor classification | Wake detection via edge and gradient analysis across frames; Kelvin wake angle used to estimate heading and approximate speed | Per-vessel activity flag with timestamp, estimated heading and speed range; GIS point layer |
| Aircraft movement detection at airfields | Frame-differencing against static runway background; persistent bright-point tracking across frames | Alert with timestamp and estimated aircraft position at each detected frame; exportable to common airfield monitoring formats |
| Construction or loading activity indicator | Aggregate motion density map across clip duration; high-motion zones flagged against a baseline clip from the same site | Site activity score (low / moderate / high) with annotated frame grabs; weekly trend report |
| Change-triggered video tasking | Very-high-resolution still imagery (covered on a sibling page) used to detect structural change; video tasking triggered only when change is confirmed, reducing unnecessary clip orders | Tasking recommendation with site coordinates and priority tier; integrated into client tasking workflow |
| Multi-site activity dashboard | Automated pipeline ingesting clips from multiple sites; standardised alerting thresholds set per site based on baseline activity characterisation | Daily dashboard feed showing per-site activity status; exception alerts pushed to client system via API |
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.