Spatial intelligence · Research
We’re researching how separate recordings can become one scene you can question, while keeping a clear line between what the evidence shows, what a model infers and what is only a possibility.
See it on your case ↗Four research directions
Several photos or camera views can be checked against their sources. A scene inferred from one image, or from a written description, leaves far more unknown. We treat these as different things, never as interchangeable evidence.
Not only what appears in a frame, but where people and objects are relative to each other, and whether that changes from one recording to the next.
Test competing accounts of where something was or what happened next, without presenting any of them as fact.
Inspect the scene from any viewpoint, with every feature linked back to the recording it came from. This is a research goal, not something we offer today.
Shown, inferred or possible
A spatial result is only as strong as its link to the record. We label each one and keep the three kinds apart.
Comes directly from a recording or photo, and can be checked against it.
Estimated from the evidence, with the method and its uncertainty stated.
One account among several, for testing. Never presented as fact.
Before any legal use
Until a reconstruction can answer all four, we present it as research, not as evidence.
The questions we are investigating
We’re researching how recordings can describe a shared space: where a camera was, what blocked its view, and how a sequence unfolded.
Study relationships between recordings with different clocks.
Investigate camera position, movement and occlusion.
Separate measured geometry from estimates and unresolved positions.
Measured reconstruction requires calibration and supporting evidence. It is not available for every recording or matter.
From recordings to geometry
Identify the original frames, camera information, source timing and any reliable measurements available.
Study image correspondences, camera poses and scene geometry. Test estimates against observations withheld from the fit.
Carry the source, method, scale and uncertainty into the result. Leave unresolved geometry unresolved.
What a model can help ask
A model can help explore perspective and competing explanations. The strength of its answer depends on the evidence that fixes the geometry.
Three different things
Authored geometry that explains a question. Dimensions and positions may be assumed.
Geometry estimated from image observations. It still needs scale, calibration and validation.
A reconstruction supported by measurements and an explicit account of error and uncertainty.
Questions, answered
No. Spatial intelligence is research in development. Some recordings do not provide enough stable visual information for a reliable reconstruction.
No. Photogrammetry estimates geometry from corresponding image observations. An animation can be authored without solving geometry from the evidence.
Only when the relevant geometry, scale and timing are sufficiently supported and validated. An illustrative model cannot establish these measurements.
Your record. Your next question.
Tell us what you’re working through.
Start a conversation ↗