Reconstructions
A reconstruction in HemoVision is a geometrically accurate, to-scale 3D representation of a documented region. It is not, however, a full photorealistic 3D model of the entire room. It is a marker-based geometric framework that supports analysis.
How Reconstruction Works
HemoVision uses a marker-guided Structure-from-Motion pipeline.
From overlapping overview images, the software:
- Detects image features and marker corners
- Matches them across photographs
- Estimates camera positions
- Triangulates marker locations in 3D
- Scales the reconstruction using known marker size
Once markers are reconstructed:
- Perspective distortion is corrected from detail images
- Detail images are aligned in 3D
After reconstruction, further analysis (stain analysis, pattern classification, trajectory analysis, ...) can begin. The reconstruction forms the geometric backbone of all further processing.
Info
For a more in-depth explanation of HemoVision's reconstruction framework, see this page.
Dividing a Scene into Reconstructions
A reconstruction represents one geometrically connected documented region. In simple terms, a reconstruction is:
One continuous spatial zone that can be captured using overlapping overview images.
When is one reconstruction sufficient?
If you can:
- Walk around the area in a semi-circle
- Capture 20+ overlapping overview images
- Maintain marker visibility across images
Then the area can likely be documented as one reconstruction.
The image below illustrates this example, where we want to capture the corner of the room. The red line indicates the movement of the photographer, whereas the triangles indicate camera positions. The shaded area indicates the region we are capturing and reconstructing.
When to create multiple reconstructions?
Create separate reconstructions when:
- You want to document regions that lie opposite of eachother
- Walking around your scene in a semi-circle is impossible
- Making overlapping overview images becomes impractical
- Marker visibility cannot be maintained across the full scene
In the left example below, capturing both the top-left and bottom-right corner of the room is impractical. Just rotating in place and taking photographs in a circle is not good practice, and does not provide sufficient variation to guarantee accurate reconstructions. Instead, divide the scene into two reconstructions, and capture each reconstruction as indicated in the example of the right below.
Below is another example. Documenting the scene in one go, as illustrated in the top sketch, creates a void where no markers are present between the two documented regions. This will hinder accurate reconstruction of the whole scene. In this case, it is better to split the scene into two separate reconstructions, as shown in the bottom sketch.
Summary
When it becomes difficult to capture your scene in one continuous motion with sufficient photograph overlap and marker coverage, it is best to split your scene into multiple reconstructions. When creating multiple reconstructions, make sure that:
- Each reconstruction has its own overview marker
- Detail marker IDs are unique across the whole scene
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