3D Reconstruction from Multiple Surfaces Using Marker-Guided Structure-from-Motion
Key idea
Bloodstain patterns are often distributed across multiple surfaces with different orientations, such as walls, floors, and objects.
This method reconstructs the three-dimensional geometry of these surfaces from photographs using a marker-guided Structure-from-Motion approach.
By placing reference markers in the scene, the software can accurately determine the relative position and orientation of all surfaces, allowing bloodstains from different surfaces to be analyzed together in a single 3D reconstruction.
Why This Method Was Developed
Traditional Area of Origin calculation methods have some important limitations.
The tangent method, for example, requires bloodstains to be on the same vertical wall. When bloodstains are scattered across different surfaces, the method becomes impractical.
Other software solutions require you to measure the location of each spatter stain at the scene itself. Again, this becomes impractical when confronted with complex patterns.
Digital approaches using 3D scanning hardware can help with measuring the locations of bloodstains. However, these methods require specialized and expensive equipment, increasing the complexity of field documentation.
HemoVision provides a middle-ground solution using a marker-guided structure-from-motion approach, eliminating the need for:
- Expensive 3D scanners
- Manual measurement
- On-scene stain selection
This article explains how HemoVision achieves a high degree of accuracy coupled with incredible ease-of-use.
How the Method Works
The method combines photography, marker detection, and 3D reconstruction into a single workflow.
Step 1 – Scene Documentation
The scene is documented using:
- Fiducial markers:
- Standard 2D photographs
Manual measurements are not needed, or kept to an absolute minimum (for more information, please see our scene documentation pages).
Step 2 – 3D Reconstruction
HemoVision uses a technique called Structure from Motion (SfM) to reconstruct the scene in 3D.
In simple terms:
- The software detects recognizable points across multiple images
- It matches these points between images
- It reconstructs their position in 3D space
To improve robustness:
- Marker keypoints are added to the reconstruction
- This ensures reliable reconstruction even on low-texture surfaces (e.g. plain walls)
- Markers also provide absolute scale, so the reconstruction can be scaled to real-world units.
The result is a to-scale 3D reconstruction of the scene. This is illustrated in the figure below (a).
Step 3 – Positioning Detail Images
Detail images (close-ups of stains) are then:
- Corrected so that any perspective distortion is removed
- Automatically scaled to real-world units
- Aligned in the 3D reconstruction using the markers visible in those images
Each image represents a locally planar surface in 3D space. This is illustrated in the figure above (b).
Step 4 – Stain Analysis and Area of Origin Estimation
Selected stains are analyzed using HemoVision:
- Impact angles and directional angles are calculated automatically
- Each stain is converted into a trajectory in 3D space
Finally, all trajectories are combined and mathematical optimization determines where they converge, resulting in an Area of Origin. To learn more about our AO-calculation method, see this article.
A visualization of the trajectories and AO is shown in the figure above (d).
Experimental Validation
To validate the accuracy of the reconstruction method, controlled experiments were performed.
Scene Setup
- Five scenes were created
- Each scene contained multiple surfaces (walls, tables, objects)
- Surfaces were randomly oriented
- Impact patterns were generated using a controlled device
Each scene was carefully designed so that the true Area of Origin was known.
Reconstructions in HemoVision
The scenes were documented using standard HemoVision guidelines. We documented each scene using both a Nikon D5500 digital camera and a Samsung Galaxy S20 smartphone camera. This way, we could also compare the reconstruction accuracy and AO accuracy between digital cameras and smartphone cameras. Moreover, we also compared accuracy between different image resolutions.
The table below summarizes some statistics about each scene.
Ground Truth Measurement
To evaluate reconstruction and AO accuracy, each scene was recorded using a 3D scanner (DotProduct DPI10). The scanner data was then used to determine:
- The true Area of Origin location, to compare to estimated AO locations.
- The true positions of markers, to compare to reconstructed marker locations.
Accuracy and Expected Performance
Reconstruction Accuracy
We evaluated reconstruction accuracy in two ways.
First, we evaluated the position error of the markers. In other words, we calculated the translational error between the ground truth marker points and the reconstructed marker points. Smaller errors indicate better reconstructions.
In total, 600 datapoints were used per combination of camera device and image resolution. The image below plots the errors for all setups. The overall median error was just below 3 mm, and no statistical difference was found between any two setups.
Second, we evaluated the rotation error of the markers. In other words, we calculated the angular difference between the ground truth markers and the reconstructed markers. Smaller angles indicate better reconstructions.
In total, 50 datapoints were used per combination of camera device and image resolution. The image below plots the errors for all setups. The overall median error was just below 0.2°, and again no statistical difference was found between any two setups.
Area of Origin Accuracy
Finally, we compared the calculated AO positions to the known ground truth locations. Averaged over the five scenes, we get:
- Average total error: ~8 cm
- Horizontal error: ~2–3 cm
- Vertical error: ~7 cm
The vertical error is typically larger because trajectories are modeled as straight lines instead of curved ones.
Despite this, the overall accuracy is well within the range reported in forensic literature.
Practical Takeaways
- The method allows analysis of complex multi-surface scenes
- Only standard photographs are required
- No need for specialized 3D hardware
- Minimal manual input is required
- Accuracy is sufficient for forensic interpretation
Scientific Reference
The reconstruction method described on this page is based on the following peer-reviewed publication:
Joris, P., et al. (2024).
Area of origin estimation from multiple arbitrarily oriented surfaces using marker-guided structure-from-motion.
Forensic Science International.
This publication describes the marker-guided Structure-from-Motion framework used to reconstruct bloodstain scenes containing multiple surfaces. The paper explains the reconstruction pipeline, the use of fiducial markers to improve geometric accuracy, and validation experiments demonstrating how the approach enables reliable Area of Origin estimation across multiple surfaces.
Readers interested in the full methodology, mathematical formulation, and experimental validation are encouraged to consult the original publication.
Found a mistake?
Did you find a mistake, or does a page contain outdated information or something that is not working correctly? Please let us know so we can review it and fix it as soon as possible.
Forentrics - Bringing Evidence into Focus