Start with the data your application needs

“Motion capture” covers very different tasks. Character animation depends on joint positions and skeleton quality. A virtual tool needs the coordinates and orientation of a specific object. Research requires repeatability and a documented measurement error. Before comparing vendors, define the result the system must deliver.

Answer five questions:

  1. What is being measured: a body, hands, fingers, a tool, a probe, a camera, or several objects at once?
  2. What output is required: XYZ position, orientation, full 6DoF, a skeleton, joint angles, or contact events?
  3. Where is the coordinate origin: the room, the body, a manikin, a field generator, or another object?
  4. What must happen in real time: recording only, streaming to an engine, simulator control, or user feedback?
  5. Which error breaks the task: stationary jitter, global position drift, trajectory gaps, image latency, or incorrect orientation?

If a simulator needs a stable virtual imaging plane, a generic “tracker accuracy” figure is not enough. Position error, orientation error and end-to-end latency to the displayed image must be measured separately.

How the three methods differ

Method Source measurement Main strength Main limitation What to test on your bench
Marker-based optical Marker positions in images from several cameras Direct 3D reconstruction in a shared coordinate system Occlusion, visibility, camera layout and calibration Trajectory continuity during real movements, not only at the centre of the volume
Markerless optical Video and a pose-estimation model No markers attached to the performer Hidden body parts must be inferred by the model Errors with your clothing, lighting, viewpoints and occlusions
Inertial Gyroscopes, accelerometers and, where present, magnetometers in IMUs Portability and independence from cameras Global position accumulates error without an external reference Drift, sensor movement on the body, recalibration and behaviour in the magnetic environment
Magnetic-inertial Sensor signals within a generated electromagnetic field plus IMU data 6DoF without direct optical visibility of the sensor Working volume, field geometry, interference and integration design Volume mapping, nearby metal and electronics, jitter, latency and recovery

Optical tracking is not one uniform technology. A marker-based system triangulates observed points; a markerless system estimates pose from images. Their errors and site requirements differ. Likewise, “inertial” and “magnetic-inertial” are not synonyms: the latter adds an external field source and a measurement chain that relates the sensor to a working volume.

What 6DoF means and why orientation alone is often insufficient

Six degrees of freedom comprise three position coordinates and three orientation parameters. For a virtual tool or training ultrasound probe, the application needs more than tilt: it must know where the working element is relative to the manikin or another object.

An inertial sensor is useful for estimating changes in orientation, but by itself it does not provide a stable global position reference. Integrating acceleration accumulates error. Before selecting an IMU-only solution, establish whether position relative to an arbitrary starting point is acceptable and how often the system can be re-referenced.

Magnetic-inertial and optical systems establish that reference differently. One uses a known field configuration; the other uses camera observations. A 6DoF label in a datasheet does not imply identical stability at the edge of the working volume, during occlusion, or next to the product’s own structure.

Occlusion and interference: compare real operating conditions

For optical 3D reconstruction, a marker must be observed simultaneously by multiple cameras. The performer’s body, tools, furniture and other people can hide it from several viewpoints at once. An additional camera helps only when it contributes an independent view rather than duplicating a neighbouring one.

Magnetic-inertial tracking does not need line of sight, but it is not immune to the environment. Metal structures, electronics, field sources and the product itself can affect measurements. Their influence depends on the actual geometry and must be tested in the assembled device.

Inertial sensors need no external optical infrastructure, yet mounting can affect the result as much as the algorithm. A module shifting on a glove or strap appears in the data as segment motion. When comparing systems, keep the mounting method fixed and repeat the same action sequence.

Latency: define the two measurement points

Sensor update rate, camera frame rate, 3D output rate and display refresh rate are different characteristics. None of them alone describes end-to-end latency.

For an interactive system, measure the interval from physical movement to the required result: a changed 6DoF stream, a moved virtual object, or a completed image frame. Record the following in the protocol:

  • whether filtering and prediction are enabled;
  • the transmission mode and update rate;
  • firmware, SDK and application versions;
  • whether the result is a mean, median, 95th percentile or maximum;
  • how frames with lost observations are handled.

A mean without long pauses can conceal exactly the problem a user experiences as a jump or a loss of contact with the virtual object.

When a hybrid architecture is justified

A hybrid is useful when one channel compensates for an observed limitation of another. For example, optical tracking can restore the position reference while a magnetic-inertial channel maintains measurement during optical occlusion. Simply connecting two systems at the same time does not achieve this.

A working hybrid architecture needs shared coordinates, time synchronisation, a consistent object model and source-selection rules. The user and application should be able to distinguish at least three states: measured, reconstructed or predicted, and lost. A smooth trajectory must not hide transitions between them.

Test not only quality during occlusion but also the return to independent measurement: whether a jump occurs, how quickly accumulated error disappears, and whether orientation remains continuous.

A comparative test programme

Run every candidate through the same route. Where possible, record source measurements and the application output synchronously.

  1. Stationary behaviour. Hold the object at several points in the working volume. Assess position and orientation jitter, not just the mean coordinate.
  2. Slow movement. Perform actions typical of precise work. Test small movements and smoothness without aggressive smoothing.
  3. Fast movement. Repeat the working accelerations and rotations. Mark losses and recovery separately.
  4. Occlusion. For optical systems, hide markers with the body and typical objects. For other methods, reproduce equivalent operating conditions and relative object positions.
  5. Environmental changes. Add real metal parts, electronics, reflective surfaces or lighting, depending on the method.
  6. Volume boundaries. Repeat the route at the centre, edges and different heights. Do not extrapolate a centre result to the entire volume.
  7. Extended operation. Record drift, heating, communication gaps and recalibration requirements.
  8. End-to-end workflow. Assess not only coordinates but also the final result: skeleton, simulator action or virtual imaging plane.

Define acceptable values and their calculation before the test. Otherwise, the comparison becomes a choice between demonstrations rather than measurement results.

Practical selection by scenario

Animation, biomechanics and a shared coordinate scene. Start by evaluating an optical system if the site and visibility can be arranged. Separately calculate cameras, viewpoints and occlusion.

XR training and hand actions in a confined space. Magnetic-inertial tracking can simplify operation without cameras, but the working volume, mounting, small movements and interference from actual equipment need testing.

Mobile capture and changing locations. An inertial system is convenient to transport. The decisive questions are acceptable global position error and recalibration frequency.

An OEM product or training ultrasound simulator. Compare assembled probe, manikin and software chains rather than isolated sensors. See the guide to 6DoF ultrasound probe tracking.

What to include in a technical brief

  • room or product dimensions and a working-volume plan;
  • the number of people and objects and their mutual occlusions;
  • a movement list identifying slow, fast and precision sections;
  • required coordinates, orientation, skeleton and output rate;
  • acceptable error and its verification method;
  • latency requirements and the final application;
  • information about metal, electronics, lighting and reflective surfaces;
  • cable, sensor mass, power and mounting constraints.

This brief allows solutions to be compared by outcome rather than by the number of claims in a marketing table.

Frequently asked questions

Is optical tracking always more accurate?

There is no universal answer without a metric and operating conditions. Optical systems directly measure observed marker positions, but occlusion and poor camera placement degrade the result. Another method may be more useful in a particular environment. Compare errors on the same route.

Can inertial drift be eliminated completely?

Do not base a design on that promise. Global position needs an external reference or another measurement channel. Test error accumulation over the duration of your workflow and recovery after recalibration.

Does no line-of-sight requirement mean no interference?

No. A magnetic-inertial system does not depend on optical visibility, but its results must be checked in the product’s real electromagnetic and mechanical configuration.

Is datasheet accuracy enough for a comparison?

No. A datasheet helps narrow the candidate list. Selection also requires measurement conditions, source data, edge-of-volume behaviour, recovery after losses and the end-to-end result in your application.

Sources and further reading