What is Site Reconstruction?
Module Overview
This module frames the course. You'll learn what site reconstruction is, where it sits in the wider reconstruction workflow, and — most importantly — the two anchors every site reconstruction depends on: a georeferenced world (satellite imagery at real-world scale) and a matched camera (a Blender camera that replicates the original photo). Everything in the course is a step toward establishing and combining those two anchors.
Introduction
Site reconstruction is the practice of rebuilding a physical location in 3D from imagery you did not shoot yourself: satellite views, street-level photos, drone video, social media stills. In a visual investigation context, the goal is rarely "a pretty 3D model." It is a verifiable, measurable 3D environment that lets you answer questions like:
- Could the witness at that location have seen the event?
- How large is the compound, the container yard, the vehicle — really?
- Does this photo match this claimed location, or is something off?
- Where was this shot taken from?
The method has been developed and tested through a live 8-week curriculum (an Oxford + Dutch Police pilot). The version in this course is the generalizable core of that curriculum, built around a deliberately generic scenario: reconstructing a house from satellite imagery and multiple photos. The house is not the point — the pipeline is.
The Two Anchors
Every site reconstruction stands on two things. If either one is missing or weak, everything built on top of it is suspect.
Anchor 1: A Georeferenced World
Satellite imagery gives you the site at real-world scale and position. In Blender, that means a scene where 1 unit = 1 meter, the origin sits at the site's real coordinates, and the ground truth of "what is where" is fixed before you model anything.
This anchor is established in Module 02 using tools like BlenderGIS and QGIS.
Anchor 2: A Matched Camera
OSINT photos give you the viewpoint: the exact position, orientation, and focal length of the camera that captured them. Camera matching (a.k.a. perspective solving, the "fSpy process") recovers those camera parameters from a single photo using the vanishing points of parallel lines in the image.
This anchor is established in Module 03.
Why Both?
- A georeferenced world without a matched camera gives you a site you can measure but cannot prove your source photo came from.
- A matched camera without a georeferenced world gives you a photo-accurate scene in arbitrary, unscaled units — impressive, but unverifiable.
Together they let you say: "This is the site at real-world scale, and this is the exact camera from which the evidence photo was taken — here is what that camera can and cannot see."
Why This Matters for Visual Investigation
The Opportunity
OSINT imagery is abundant and often geolocatable. Site reconstruction turns flat images into a measurable, explorable environment — the difference between looking at evidence and interrogating it.
The Caveat
Reconstruction is interpretation, not recording. A reconstructed site is a model of a site, built from assumptions (imagery resolution, matching accuracy, what you chose to model). It must be documented and presented with its uncertainties — which is exactly what Module 06 covers.
Where It Fits in the Workflow
Site reconstruction is one branch of geometric reconstruction. The KB's geometric reconstruction playbook distinguishes between object-level, site-level (this course), and environmental reconstruction. The camera matching technique page is the core skill this course operationalizes.
Key Concepts at a Glance
| Concept | Definition |
|---|---|
| Georeferenced baseline | A Blender scene aligned to real-world coordinates at 1:1 scale, built from satellite imagery |
| 1:1 scale | The convention that 1 Blender unit = 1 meter, so all measurements are in real-world units |
| Vanishing point | The point where parallel lines (not parallel to the image plane) converge in a perspective image |
| Perspective solving (camera matching) | Recovering camera position, orientation, and focal length from vanishing points in a single image |
| Photomatch | The verification step: comparing a render from your matched camera against the original photo |
| Scale hierarchy | The ranked source-reliability system for scale references (detailed in Scale Estimation) |
Knowledge Check
- What are the two anchors of a site reconstruction, and what does each one give you?
- Why is a matched camera useless for measurement unless the scene is at 1:1 scale?
- Name one question an investigation can answer with a reconstructed site that it cannot answer from the flat photo alone.
Check your answers
- The georeferenced world (real scale + position from satellite imagery) and the matched camera (the exact viewpoint of the source photo). One gives you measurability, the other gives you provenance of the viewpoint.
- Camera matching recovers relative geometry, not absolute scale — the recovered camera is only in scene units. Without a 1:1 georeferenced scene, those units are arbitrary, so any distance you measure is meaningless.
- Examples: line-of-sight from a specific spot, the true size of a structure, the camera height at capture, whether an object was present in a scene reconstructed from another angle.
Exercise: When to Reconstruct a Site
For each scenario, answer:
- Is site reconstruction the right tool here? (Yes / No / Only partially)
- Which anchor do you establish first — the georeferenced world, or the matched camera? Why?
- What is your first concrete step in Blender or a GIS tool?
Scenario A
You have a drone video orbiting a two-story house and its compound, suspected of being used as a weapons cache. You want to measure the compound, test sightlines from neighboring buildings, and later show where a specific van was positioned.
Scenario B
A single street-level photo shows a suspect's car parked in front of a building. You want to determine whether a witness standing 200 m away could have read the license plate.
Scenario C
A social media post claims to show a building destroyed by a strike in a named city district. You want to check whether the visible skyline and street layout actually match that district.
There are no single "correct" answers — this is a judgment exercise. Compare your reasoning against the Camera Matching reference module (in the sidebar) and the Analyzing Footage page. You'll put these judgments into practice in the Final Project.
Summary
Site reconstruction rebuilds a location in 3D from satellite and OSINT imagery. It stands on two anchors: a georeferenced 1:1 world (satellite baseline) and a matched camera (perspective solving of the source photo). The course proceeds from anchor to anchor to model: baseline → camera → model → verify → document.
Resources
Knowledge Base Pages
- Camera Matching — the core technique this course operationalizes
- Camera and Perspective — the foundational perspective concepts
- Geometric Reconstruction — the playbook: three scopes and the full phase flow
- Measurements, Scale & Dimension — the scale hierarchy you'll lean on constantly
Add-ons
- fSpy — vanishing-point camera estimation, imported into Blender
- Perspective Plotter — real-time camera matching inside Blender
Upcoming Videos in This Course
| Module | Video |
|---|---|
| 02 | Establishing a Measurement Baseline from Satellite Imagery |
| 03 | Applied Camera Matching from Photos |
| 04 | Modeling from Matched Photos |