PoseIQ Decision & Measurement Assurance Framework
Validation is not one universal accuracy number. Every movement decision deserves the right level of evidence for its measurement target, capture setup, and risk level.
PoseIQ starts with the decision being made, then asks what measurement evidence is required. A movement game, a coaching trend, a range-of-motion measure, and a clinical workflow do not need the same proof.
Validate the decision, not a vague AI label
PoseIQ treats validation as a chain: intended use, movement target, camera setup, measurement geometry, evidence required, and use status. If any part of that chain is weak, the decision should be limited, conditional, or unsupported.
Hold still inside the dashed box
This demo estimates frame-to-frame video change. It shows why lighting, camera position, movement, and jitter affect measurement quality. It is not a body-tracking validation test.
The score reflects video stability only. It is a setup-quality teaching tool, not a measure of movement accuracy.
Match evidence to the intended decision
This explainer estimates use status from the decision, measurement geometry, capture setup, and risk level. It is not a certification tool.
Use status
From use case to use status
Evidence scales with risk
| Application | Risk | Evidence required |
|---|---|---|
| Interaction or game | Very low | Functional test and false-trigger review |
| Rep counting | Low | Repeatability and manual video audit |
| Balance or posture trend | Moderate | Setup guardrails and repeatability checks |
| Range of motion | High | Comparison with an appropriate reference |
| Diagnosis or treatment | Unsupported | Not supported by this framework |
Camera geometry changes the claim
| Measurement | Use status | Guardrail |
|---|---|---|
| Timing and rhythm | Suitable with checks | Useful in accessible setups. |
| Sagittal flexion / extension | Conditional | Camera should be perpendicular to movement. |
| Frontal plane movement | Limited | Sensitive to perspective and alignment. |
| Transverse rotation | Guardrail | Needs multi-view or reference validation. |
Current assurance levels
No vendor-specific model claims
| Route | Best fit | Constraint |
|---|---|---|
| 2D landmark backbone | Posture, timing, ROM trends | Depth ambiguity |
| Real-time detector route | Interaction, rep counting, fast feedback | Fine movement noise |
| Multi-view fusion | Cross-plane movement and 3D trajectories | Requires controlled setup |
| Reference lab setup | Benchmarking and high-risk evidence | Not a consumer workflow |
Evidence is task-specific
Markerless movement analysis is strongest when the intended decision, camera setup, movement plane, and reference method are clearly defined. Single-camera systems can be useful for accessible workflows, but results should be interpreted within the limits of the setup.
- Systematic reviews of markerless gait analysis generally report stronger performance for many timing measures than for some joint rotations.
- Comparisons between markerless and marker-based systems support task-specific and plane-specific interpretation.
- Low-cost camera workflows can support interaction, education, and limited screening when evidence and boundaries are clearly stated.
- Framework publication: Mokhtarzadeh, H. (2026). Accuracy is not one number: A measurement assurance framework for camera-based human movement. engrXiv. https://doi.org/10.31224/8260
PoseIQ does not claim one universal movement accuracy score. Validation depends on the intended decision, measurement target, capture setup, reference method, and risk level. Where evidence is insufficient, use status should remain conditional, limited, not recommended, or unsupported.
This is a living research and engineering framework. It is not a clinical device, certification, or diagnostic system.