Why “one camera per N square metres” does not work
Two studios with the same floor area can require different configurations. In one, a performer walks around an open circle. In another, they sit on the floor, work with their hands behind equipment, or interact with a second performer. The second scenario creates longer and more frequent occlusions even though the area is unchanged.
Floor area is only one of the inputs:
| Input | What to record | What it affects |
|---|---|---|
| Working volume | Length, width, movement height and clearance to walls and barriers | Field of view, distances and overlapping sectors |
| Movements | Walking, turning, squatting, floor work and fast gestures | Required camera elevations and occlusion duration |
| Performers and objects | Number of people, tools, furniture and equipment | Mutual and static blind spots |
| Markers | Body or object layout, size, active or passive type | Visibility, distinguishability and usable distance |
| Camera and lens | Resolution, field of view, frame rate, focus and actual operating range | Coverage and observation quality |
| Required output | Individual points, rigid bodies, skeleton, fingers or real-time stream | Acceptable gaps and processing requirements |
Camera count is therefore an outcome of geometry and acceptance criteria, not the first number to place in an estimate.
The triangulation minimum is not a working configuration
Calculating a point’s 3D position requires synchronised observations from at least two cameras. At that minimum, losing either view means losing the point. Even when two cameras see the marker, nearly identical viewing directions provide less favourable geometry than separated viewpoints.
Third and subsequent independent views are useful for specific reasons:
- preserving reconstruction when one view is blocked;
- seeing markers on opposite sides of a body or object;
- separating observations by elevation during low poses;
- improving triangulation geometry;
- reducing dependence on one part of the working volume.
A camera added next to an existing one may change very little: both can experience the same occlusion. Count not just cameras, but useful independent observations during every critical movement phase.
How to find blind spots before purchasing
Draw a site plan and mark only the area where data is actually required, rather than the whole room. Add obstacles and height levels. A single horizontal slice at waist height is insufficient if the performer raises their arms or lies on the floor.
Common causes of data loss
Body self-occlusion. During a turn, chest markers are hidden from cameras behind the performer; during a bend, neighbouring body parts obscure one another. Views from different sides and elevations are needed.
A second performer or object. A person, desk, simulator or tool can block several cameras at once if those cameras are clustered together.
The edge of the field of view. A marker is inside the room but outside the image. Test the entire volume, including its upper and lower boundaries.
Insufficient distance or image size. A camera covers the area, but the marker is too small or unstable in the image. Coverage on a drawing is not the same as a reliable observation.
Stray reflections and sources. For IR-illuminated systems, inspect glare and reflective objects in the real scene. A simple visibility map does not describe their effect.
How to place cameras around the perimeter
For typical body or rigid-object capture, cameras are distributed around the working volume and aimed inward. This increases the chance of observing different sides of the target. The optimal layout depends on the application, but several principles prevent common mistakes.
- Separate the viewpoints. Do not place every camera on one wall if the target turns through different orientations.
- Use different elevations where needed. High cameras cover the general volume; lower views help with floor work and low markers. A low level is not universally required—the scenario determines it.
- Model the actual field of view. Use the specific lens parameters and sensor orientation on the plan, not a generic sector.
- Check overlapping fields of view. Areas observed by several cameras matter more than the sum of disconnected sectors.
- Include mounting and servicing. An accidental shift of a camera or truss changes calibration. Leave safe access for adjustment.
- Do not confuse a demonstration with a datasheet. A simplified 3D model helps identify weak layouts but does not replace testing the actual camera and markers.
OptiTrack’s placement guide separately recommends peripheral layouts, diverse viewpoints and camera separation: closely spaced devices capture similar images and provide less help during occlusion.
Step-by-step configuration planning
1. Write the acceptance scenario
Describe 30–60 seconds of movement that must be captured. Include the least favourable actions: turning away, bending, squatting, hands close to the body, approaching the boundary and interacting with an object.
2. Define the working volume
Mark only the required area and specify minimum and maximum marker heights. Separately identify areas where data is critical and technical walkways where capture is unnecessary.
3. Create an initial layout
Start with cameras around the perimeter. For each one, check distance, angle, elevation and visible area using the documentation for the specific model. Do not apply one template to every lens and operating mode.
4. Check independent observations
At every critical route point, assess not only the presence of two observations but also their separation. Mark locations where one obstacle blocks several cameras simultaneously.
5. Add or relocate cameras for a stated reason
If a shared wall causes the problem, relocation may help more than another nearby camera. If the floor causes occlusion, change elevation or add a lower view. If the marker projection is too small, distance, lens or camera model may need to change rather than camera count alone.
6. Record and measure the gaps
A geometric prototype selects promising layouts. The final decision comes after recording the real route with the selected markers, suit, lighting and software processing.
A test route for coverage validation
| Stage | Action | Data to retain |
|---|---|---|
| Static poses | Stop at the centre, edges and in a low pose | Jitter and marker-identification instability |
| Rotations | Turn slowly through 360° with arms down and raised | Gaps by marker and gap duration |
| Low work | Squat, bend and perform movements near the floor | Visibility changes with height |
| Fast gesture | Repeat the actual fast working movement | Losses, association errors and recovery |
| Occlusion | Pass behind an object or interact with another person | Number of simultaneously lost views |
| Boundary | Walk around the working-volume perimeter | Areas leaving the field of view |
| Return | Re-enter a clearly visible area after a loss | Time and correctness of trajectory recovery |
For each stage, retain not only the final skeleton but also source observations or visibility diagnostics where the system provides them. A visually smooth reconstructed skeleton can hide a long measurement gap.
Metrics to include in acceptance
- the proportion of time with valid reconstruction of key markers or the rigid body;
- the number of gaps and the distribution of their durations;
- separate results for the centre, edges and different heights;
- behaviour when several markers are occluded at once;
- time to restore correct identification;
- stationary marker jitter;
- error at reference points where a reference standard is defined;
- 3D output rate and end-to-end application latency as separate metrics.
Set thresholds before comparing layouts. Previsualisation can tolerate longer reconstruction than simulator control or a research protocol. A single “coverage quality” percentage without a definition does not replace these measures.
Three examples where the task changes, not the area
One performer in a 6 × 6 m studio. Test a perimeter layout, rotations and approaches to the boundary. If all movements are standing and occlusions are simple, low viewpoints may not be necessary.
Hand and floor work. Add low poses, hands occluded by the body and small markers to the route. Elevation, distance and resolution may matter more than room area.
Two performers and equipment. Test mutual occlusion and the simultaneous loss of several views. Cameras may need to be redistributed and the working volume divided into sectors for analysis.
For a preliminary TAU Lumen estimate, use the plan, height, number of people and movement description. Determine the tracking method before a detailed estimate: not every task requires an optical system.
Frequently asked questions
Two cameras see the marker. Is that enough?
Only while both observations and suitable geometry are maintained. If one view is occluded, reconstruction of that point stops. A working system needs observation reserves throughout the scenario.
Are eight cameras always better than four?
No. Eight cameras with similar viewpoints can reproduce the same blind spots. Four well-separated cameras may provide more useful observations. Compare specific layouts on the same route.
Can a system be designed from the floor plan alone?
A plan supports a preliminary configuration, but it does not fully describe movement, markers, reflections or processing. An acceptance scenario must be recorded before final approval.
Why not promise a camera count immediately?
Area alone does not describe height, lenses, distances, marker layout or occlusion. A responsible calculation states its assumptions and is then confirmed by testing.
Sources and further reading
- OptiTrack — Camera Placement: the two-camera minimum, peripheral layouts, elevations and independent viewpoints.
- OpenMoCap — Rethinking Optical Motion Capture under Real-world Occlusion: why real occlusions can be prolonged and affect neighbouring markers together.
- Vicon — Types of Motion Capture: the role of optical tracking among other methods and the effect of workflow on selection.