# Geometric Reconstruction

## Introduction

Geometric reconstruction is the end-to-end practice of rebuilding a scene from visual evidence in Blender. It operates at three scopes:

- **Object-level:** Reconstructing and measuring individual objects from photos — anchored by the [scale pillar](1-foundation-preparation/scale-estimation.md) (the source-reliability hierarchy and error propagation).
- **Site-level:** Rebuilding a location in 3D from satellite and OSINT imagery — a georeferenced 1:1 baseline plus a matched camera. Operationalized in the [Site Reconstruction course](../courses/site-reconstruction/syllabus.md).
- **Environmental:** Rebuilding the geospatial context — terrain, coordinate systems, and ground control points — the "Ground Truth" stage on which the investigation plays out.

The phases below cover one full pass. Phases 1 and 5 (data handling and reporting) are owned by the [Foundation course's Data lifecycle](../courses/foundation/03-data-and-methodology/index.md); Phases 2–4 are detailed on this page.

## Phase 1: Ingestion & Processing

The data-handling stages are covered by the **Data lifecycle** — a single, unidirectional path:

1. [Collect](../courses/foundation/03-data-and-methodology/collect.md) — acquire and document sources (registry, provenance, authenticity, technical assessment).
2. [Preserve & Store](../courses/foundation/03-data-and-methodology/preserve-store.md) — sanctuary for originals, hashing, working copies, segmentation.
3. [Process](../courses/foundation/03-data-and-methodology/process.md) — transform sources into usable assets (temporal alignment, enhancement), with every change logged.

Phase-specific reconstruction decisions continue in **Scene Decomposition** below.

## Scene Decomposition

**Asset Strategy & Branching.** Scene decomposition involves isolating, categorizing, and preparing every individual element within the frame for reconstruction. This phase turns a complex scene into a structured inventory of assets ready for object and environmental reconstruction.

### Asset Identification (Inventory)

- **Macro-Static (Infrastructure):** Buildings
- **Micro-Static (Furniture):** Street signs, benches, trees, trash cans.
- **Dynamic Entities (Movers):**
  - **Biometric:** Humans, crowds, animals (Agency).
  - **Kinematic:** Vehicles, machinery (Physics).
- **Ephemeral Events:** Explosions, smoke plumes, fire.
- **Environment:** Sun position, weather conditions (wet roads/puddles), debris.

#### Naming & Organization

Assign a unique ID to every item in the Inventory.

Don't use naming conventions that are subject to change (No `Car_Lada_Niva_`, but `Car_Sedan_White`). Pick categories, classifications or a taxonomy and visual aspects that are static (a car doesn't change from black to blue, nor does it change from Sedan to Hatchback).

**Naming Convention:** `Category_Description_ID` (e.g., `MACRO_Building_TownHall_01` or `DYN_Car_RedSedan_02`).

**Folder Structure:** Create a directory for each asset containing two sub-folders:
- `/Reference` (Blueprints, OSINT photos).
- `/Data` (Measurements, specs).

### Intelligence (Reference Gathering)

Try to find "Ground Truth" data for each item in your Inventory.

**OSINT / Specs:** Locate blueprints, satellite maps, vehicle manuals, or weather reports.

**Dimensional Verification:** Confirm the height, width, and length of objects. Example: Don't guess the height of the lamp post; find the manufacturer spec or measure it relative to a known object.

**Alternative Angles:** Find existing footage (Street View, News, Social Media) that shows the object from sides not visible in your primary evidence.

### Establish Methodology per Object

| Strategy | Definition | Best Use Case |
|----------|------------|---------------|
| **SOURCE** | Library Assets | Micro-Static & Generic Kinematic. Standard objects (Trash cans, Oak trees, generic Ford Focus) where the specific scratches/dents don't matter. Do not reinvent the wheel. |
| **GENERATE** | Automated Creation, Procedural | Macro-Static & Environment. Complex terrain, debris fields, or large city blocks. Use Photogrammetry (Drone/Google Earth) or Scatter tools (Gravel, Grass). |
| **MODEL** | Hand-Crafted | Key Evidence. The specific building where the event happened, or a unique vehicle with specific damage. These must be built vertex-by-vertex to match the reference perfectly. |
| **SIMULATE** | Physics-Based | Ephemeral Events. Explosions, fire, smoke. These cannot be modeled; they must be calculated using fluid dynamics based on the visual evidence. |

### Evidence Assessment

What source material do you have? This determines which reconstruction techniques are possible.

**Visual Evidence Density:**
- **Sparse (1-5 viewpoints):** Limited angles, use camera matching/perspective projection
- **Medium (6-30 viewpoints):** Structure-from-Motion (SfM) photogrammetry feasible
- **Dense (30+ viewpoints):** Full photogrammetry or Gaussian Splatting possible

**Visual Evidence Quality:**
- **High:** DSLR photos, stabilized video, minimal compression
- **Medium:** Smartphone footage, CCTV with moderate resolution
- **Low:** Compressed social media video, long-distance footage, heavy motion blur

**Temporal Coverage:**
- **Single moment:** Static reconstruction
- **Discrete events:** Keyframe animation between known states
- **Continuous:** Full motion tracking required

**Supplementary Data Availability:**
- **Geospatial:** GPS coordinates, satellite imagery, maps, blueprints available?
- **Technical specs:** Vehicle dimensions, building plans, object specifications?
- **OSINT:** Can you find additional angles from social media, news footage, Street View?

**→ Output:** Identify which reconstruction techniques are viable (photogrammetry vs. modeling vs. camera matching).

## Object Reconstruction (Summary)

This branch focuses on reconstructing key objects from the evidence inventory established in Scene Decomposition.

- Prioritize objects that are directly tied to the research question.
- Use reference-driven modeling, camera-constrained modeling, or photogrammetric methods based on available sources.
- Establish scale and uncertainty **before** forwarding assets to synthesis.

The full scale pillar — the five-tier source-reliability hierarchy, confidence decision matrix, error propagation, and verification logging — lives on the [Scale Estimation & Source Reliability](1-foundation-preparation/scale-estimation.md) page.

## Environmental Reconstruction

**Geospatial.** Environmental reconstruction focuses on rebuilding the geospatial context of the scene. This phase establishes the "Ground Truth" — the accurate, real-world stage upon which the investigation plays out.

### Geolocation

**Verify coordinates.** If no location is found, proceed with Relative Reconstruction (arbitrary origin).

**Chronolocation:** Determine the time of day and date to accurately simulate lighting conditions (sun position). See [Geolocation](1-foundation-preparation/geolocation.md) and [Chronolocation](1-foundation-preparation/chronolocation.md) for the supporting techniques.

### Terrain & Context Acquisition

**Vector Data:** Import OSM (OpenStreetMap) footprints via QGIS.

**Elevation:** Import DEM (Digital Elevation Models) for accurate terrain topography.

**Photorealistic Context:**
- **Satellite:** Google/Bing Satellite imagery (Regular and possibly Historical/Dated).
- **3D Tiles:** Import Photogrammetry Mesh (Google Earth 3D) or Geometry Rip (via RenderDoc).

See [Terrain & Landscape Generation](3-acquisition-generation/terrain-landscape-generation.md) for the full terrain pipeline.

**Critical Check:** When using BlenderGIS ensure CRS (Coordinate Reference System) consistency across all datasets.

### Coordinate System Verification

- **Document CRS:** Explicitly state which CRS you're using (e.g., WGS84, UTM Zone X).
- **Resolution standards:** Specify minimum acceptable resolution for satellite imagery (e.g., "prefer 1m/pixel for urban scenes").
- **Temporal matching:** If your event happened in 2023 but Google Earth imagery is from 2019, document what changed (construction, demolished buildings).
- **Vertical datum:** Clarify whether elevation is relative to sea level (EGM96, EGM2008) or ellipsoid.

### Ground Control Points (GCPs)

- **Identify Landmarks:** Find fixed landmarks visible in both your evidence AND satellite/maps (corners of buildings, painted road markings).
- **Anchor Points:** These become your "anchor points" for aligning everything.
- **Documentation:** Document GCP coordinates in a table with confidence ratings.

## Phase 5: Synthesis & Validation

The general validation and reporting is covered by **[Validate & Report](../courses/foundation/03-data-and-methodology/validate-report.md)** — assembly, spatial/geometric verification, confidence visualization, and the analytical report with honest limitations and boundary awareness.

Phase-specific assembly decisions (which terrain and asset layers to merge) follow from Scene Decomposition and the phases above.

## Documentation for Methodology

### What to document during Scene Decomposition

### In Your Method Section

"The scene was decomposed into [Number] distinct assets. Key evidence, such as [Object A], was modeled manually based on [Reference]. Contextual elements were sourced from [Library Name] or generated procedurally."

### In Your Decision Log

Record the following:
- The source of dimensions for each key asset (e.g., "Vehicle dimensions from 2018 Manufacturer Manual").
- Why a specific modeling strategy was chosen (e.g., "Used photogrammetry for the statue due to complex geometry and available drone footage").
- Any assumptions made about object scale where ground truth was unavailable.

### Verification

"Modeled assets were overlaid on source imagery to verify silhouette alignment and scale."

### What to document during Environmental Reconstruction

### In Your Method Section

"The environment was reconstructed using [CRS Name] as the coordinate reference system. Terrain data was derived from [DEM Source] with a resolution of [X] meters. Vector data for building footprints was imported from OpenStreetMap and verified against satellite imagery dated [Date]."

### In Your Decision Log

Record the following:
- The specific CRS code (e.g., EPSG:32633).
- The source and date of satellite imagery used.
- Any discrepancies found between map data and visual evidence (e.g., "Satellite image shows a building that is not present in the video evidence; confirmed demolition date via news source").
- Coordinates of Ground Control Points used for alignment.

### Verification

"The alignment of the 3D terrain model was verified by checking the position of [Landmark A] and [Landmark B] against their known GPS coordinates."

## Summary

Geometric reconstruction is a layered process: ingest and document the data, decompose the scene into an inventory with a creation strategy per asset, reconstruct objects with a defensible scale, anchor everything in geospatial ground truth, and validate the whole assembly against the source evidence. Each layer earns its keep in the final report — the methodology section is where these phases become proof.

**Key Takeaways:**
- **Inventory before building:** List everything, name it consistently, and assign a strategy per asset.
- **Ground Truth first:** Verify dimensions and geolocation with external data whenever possible.
- **Scale before synthesis:** Establish scale and uncertainty before objects move into validation.
- **CRS consistency:** All data layers must share a Coordinate Reference System.

**Further Resources:**
- [OpenStreetMap](https://www.openstreetmap.org)
- [Google Earth Pro](https://www.google.com/earth/versions/)
- [QGIS](https://qgis.org)
- [BlenderGIS Addon](https://github.com/domlysz/BlenderGIS)