Modeling Case Studies
Introduction
To illustrate what is possible in practice, this page gathers real-world investigations and reverse-engineers the likely workflow combinations behind them. The goal is not to provide definitive production breakdowns, but to give you a clear orientation point for how tools, techniques, and workflow patterns can be combined in applied work. Each case demonstrates a blend of techniques, and all feature photomatching (the process of overlaying 2D images onto 3D models). Our focus here is on the 3D modeling aspects of these projects. Note that photomatching rarely achieves a perfect 100% alignment. Some degree of approximation is inherent to the process.
These rudimentary breakdowns serve as observational studies rather than critiques or definitive methodologies.
Data Sources
Combining satellite imagery, CCTV footage, drone scans, and blueprints.
Techniques
Photogrammetry, SubD modeling, procedural generation, and simulation.
Workflows
From simple blockouts to detailed forensic reconstructions.
Case 1: Finding an RSF Military Camp
Organization: Centre for Information Resilience (CIR)
Research question: Can we link any cars or people we see [in the videos] in Libya with the fighting in Sudan?
Object of interest: Vehicle
This investigation relies heavily on the car. In such a case you can either model it yourself from reference, or source and adapt it. That's the modeling challenge. The identification proces of the car in different places relied primariy on damage, which seems to have been added by texturing (which in turn requires UV unwrapping or a high density mesh for vertex painting). The hole in the back door might be a texturing job and the structural damage on the door could have been sculpted or manualy deformed (for example by using the proportional editing tool).
- Sourced Model: The model is likely sourced. It is very detailed, meaning that very likely a SubD workflow was used to create this base model.
- Custom Details: Then military details are added in (the hatches). These pieces are probably made by hand and, to maintain a similar fidelity to the base SubD model, are made with a SubD workflow.
- Texturing & Damage: The answer to the main question relies on matching the car by its damage and texture. For the texturing, we would need to UV unwrap at least parts of the car model so that we can paint over them (or project and bake the original image onto the model). The model also features damage that goes beyond texturing. The door was likely seperated from the car body and its structural damage either sculpted or direct modeled by using (something like) the proportional edit tool.
- Bullet Holes: Modeling bullet holes is very time consuming so you are better off using a stencil.
Case 2: ‘Smoke and Lies’
Organization: Global Investigative Journalism Network (GIJN) Read the Story
Likely research question: Can we disprove the official government narrative that deaths occurred because "the key was missing? Subject: Detailed 3D model of the interior of a detention center.
Photomatch
Base Model
- Environment: Placed within a photogrammetry-like environment (very likely Google Eart/maps import) with height information on the topography.
- Interior Modeling: According to the article "Using various reference sources, including floor plans and footage from the 15 fixed cameras in the building, the team built a 3D model of the detention center."
- Simulation: Smoke simulation was used to test official accounts.
The article provides a detailed description of the modeling methodology:
Sapoch used a variety of methods to build the interior and exterior scenes. The exterior consisted mainly of “mesh” brought in from the Google Maps Tiles API (the same 3D data built into Google Earth), as well as supplemental terrain data; they modeled the interior manually by referencing floor plans and additional reference imagery from inside the building.
“This aspect was much more time-consuming, because we weren’t able to access the site ourselves and there were various inconsistencies with the floor plans that we had access to,” Sapoch says.
They had access to confidential documents from an unnamed source, and were also able to piece things together with a wealth of layout reference material. Along with the CCTV footage, they had access to post-fire walk-throughs of the building, which provided further information about specific room layouts.
“We were lucky in so far as with the access to all this material, we were able to precisely identify the placement of the CCTV cameras within the layout of the building and essentially bring those exact camera angles into Blender and project the footage as a matched background image,” he notes.
Sapoch adds: “An equally large amount of time needs to be dedicated in 3D reconstruction projects like this to more mundane aspects of the workflow — like image matching, setting lighting and materials, animating camera movements, render tests, and so on. It’s not a small effort.”
Case 3: Shireen Abu Akleh
Organization: Forensic Architecture View Investigation
Research questions: Who killed Shireen Abu Aqleh? Was the shooting intentional or accidental? Did the shooter(s) know they were targeting journalists? Were the journalists clearly identifiable as press? Could their PRESS vests be seen from the shooter's position?
Subject: The Extrajudicial Killing of a Journalist.
Photomatch
Base Model
Workflow:
- Base model is a detailed photogrammetry scan by drone and some on the ground scanning (Tree), placed on top of satellite imagery.
- Blockout models of buildings were added to the photogrammetry scene.
- Presumably matched the sun position to the time of the incident (shadow directions differ).
- Modeled or sourced military vehicles.
- Modeled or sourced additional vehicles.
- Posed and positioned human reference models in static scenarios according to timeline.
- Field of view and trajectory modeling.
Techniques:
- Static Images: Used to capture the position of actors, no animation.
- Vests: Modeled and textured.
- Foliage: Modeled or sourced the foliage. This level of detail would not have been captured by a photogrammetry scan since Photogrammetry is terrible with the intricacies of foliage.
- Car: Low level of detail.
- Bus: High level of detail, likely a sourced model (interior is visible, too time consuming to do by hand and doesn't play a role in answering the question).
- Analysis: Modeled bullet trajectories and line-of-sight.
Case 4: Reconstructing the Melilla Massacre
Organization: Lighthouse Reports View Investigation
Photomatch
Base Model
Observations:
- Repeating Parts: A lot of repeating parts.
- Barbed Wire: Difficult to model by hand, likely procedural, modular, or a combination of both.
- Fences: Same for the fences. The parts and the chain links would likely be procedural, fitting them into a single fence piece then instancing (linked copy) this single piece across the scene
- Lamp Posts: Same approach, linked instance of a base model placed across the scene many times.
- Building: Simple geometry, with either shading errors or missing faces.
- Terrain: Modeled on top of a digital elevation map (likely exported from QGIS) or modeled flat first and overlaid the DEM later.
Summary
These cases demonstrate how different modeling approaches are selected based on the investigative needs.
Key Takeaways:
- Mapping & Data Visualization: Integrating 3D models with geospatial data.
- 3D Modeling & Photogrammetry: Combining hand-modeled assets with scanned environments.
- Video Synchronization: Using 3D space to align and verify video evidence.