AI movement analysis for physical therapy: how it works.

In short: AI-powered movement analysis uses pose estimation (a form of computer vision) to track body landmarks in ordinary smartphone video and turn them into clinical metrics: gait speed, step length and symmetry, joint angles, sit-to-stand time, and postural sway. For physical therapists it replaces stopwatches, goniometers, and visual rating with consistent, recorded measurements, without markers, wearables, or a motion lab.

How AI motion analysis works

1. Capture. The patient performs a standardized test (a hallway walk, a sit-to-stand, a balance stance) while a phone or tablet records it. Guided framing keeps captures consistent from visit to visit.
2. Pose estimation. A computer-vision model locates body landmarks (joints, limbs, trunk, head) in every frame. This is the "markerless" part: the AI finds the body directly in the video, where a lab would use reflective markers. How markerless motion capture works →
3. Biomechanics. Landmark positions over time become measurements: joint angles, stride and step length, cadence, timing of each phase of a test, and how far the body sways.
4. Scoring and trending. Results are scored automatically against the test protocol and compared with prior visits, so change over the plan of care is visible and documented.

What it measures in physical therapy

• Gait: gait speed, cadence, step length, and left-right symmetry from a short walk. See the 10-meter walk test and gait speed as a vital sign.
• Functional mobility: Timed Up and Go, 30-second chair stand, and sit-to-stand timing, with each phase of the test broken out.
• Balance: static and dynamic balance and postural sway, which are hard to quantify by eye.
• Range of motion: joint angles captured the same way across raters and visits (range of motion without a goniometer).

Norms and cut-offs for these tests are collected in the functional test reference values.

Where it fits in a PT practice

• In the visit: standardized tests are captured and scored in moments, so the number is in the note before the patient leaves.
• Between visits: patients complete short, guided re-tests at home, and the therapist sees the results before the next appointment. This also supports remote therapeutic monitoring (RTM).
• In documentation: objective, time-stamped measurements show progress toward goals for payer documentation and reduce rater-to-rater variability across a clinic.

Home exercise programs: the same pose estimation can count repetitions while the patient exercises at home, so completion is measured rather than self-reported. That turns a home exercise program from a handout into adherence data the therapist can see and act on. How AI rep counting works →

Some AI tools only coach exercise; others only assess. Kinetically does both: standardized clinical assessments with objective metrics, and home programs with automatic rep counting and adherence tracking, in the same app.

How accurate is AI movement analysis?

The published research is about the method, not any one product, and it is encouraging for the measures PTs use most:

• A handheld smartphone paired with computer-vision pose estimation measured whole-body movement with mean joint-angle errors under 3° and pelvis-translation errors of a few centimeters versus laboratory motion capture — and tracked clinically meaningful change in prospective clinical use. (Peiffer JD et al., 2025)
• A systematic review of smartphone-based gait analysis in chronic health conditions reported good-to-very-good validity against established reference systems in clinical settings — with precision more variable in unsupervised home environments. (Bea T et al., 2025)
• A systematic review in the Journal of NeuroEngineering and Rehabilitation found markerless motion capture is a practical and increasingly validated alternative to marker-based laboratories for clinical measurement, while noting accuracy varies by joint, plane, and task. (Lam WWT et al., 2023)
• AI-driven markerless capture showed excellent within- and between-session reliability for spatiotemporal gait parameters and sagittal-plane kinematics in healthy adults; out-of-plane (frontal/transverse) measures were flagged for further refinement. (Schoenwether B et al., 2025)
• In stroke survivors, AI markerless motion capture showed good-to-excellent validity for gait speed, stride length, and stance time versus reference measurement — supporting its use for key spatiotemporal metrics, with some parameters (e.g., stride width) interpreted with caution. (Alammari B et al., 2025)

The evidence is strongest for spatiotemporal gait metrics and sagittal-plane joint angles. Frontal- and transverse-plane measures, step width, and unsupervised home captures are less precise. More detail is on the evidence and validation page.

Limits to keep in mind

• Capture conditions matter: framing, lighting, clothing, and camera angle affect accuracy, which is why guided capture protocols exist.
• Not every measure is equally reliable: treat out-of-plane and width-based measures with more caution than gait speed or sagittal joint angles.
• Decision support, not diagnosis: the metrics inform clinical reasoning; the therapist interprets them and decides.

AI video analysis vs. wearables vs. a motion lab

Marker-based labs remain the research reference standard but need dedicated space, equipment, and staff. Wearable sensors are strong for continuous, real-world activity but must be worn, charged, and placed correctly. Smartphone AI analysis trades a little precision on some measures for reach: any clinic, any room, any visit. The full comparison is in smartphone vs. wearables vs. lab motion capture.

How to evaluate an AI movement analysis tool

• Which standardized tests does it capture, and does it follow each test's published protocol?
• What validation evidence exists, for which metrics, and against what reference standard?
• Does it produce the same number for the same patient across two clinicians?
• Can patients re-test and exercise at home, with reps and adherence tracked automatically, and does it support documentation and RTM?
• Is it HIPAA-aligned with a BAA available, and does data export to your EHR?

The movement analysis apps buyer's guide covers the categories of tools and what to ask vendors.

How Kinetically does it

Kinetically is AI-powered movement analysis built for clinicians. It tracks 1,200+ body landmarks from ordinary smartphone or tablet video, guides the capture of standardized tests, scores them automatically, and trends every result over time, in clinic or at home. Home exercise programs use the same camera to count reps automatically and track adherence. See Kinetically for physical therapy or every measurement it captures.

References

1. Peiffer JD, Shah K, Djuraskovic I, et al. (2025). Portable Biomechanics Laboratory: Clinically Accessible Movement Analysis from a Handheld Smartphone. arXiv preprint 2507.08268. doi:10.48550/arXiv.2507.08268
2. Bea T, Chaabene H, Freitag C, et al. (2025). Psychometric Characteristics of Smartphone-Based Gait Analyses in Chronic Health Conditions: A Systematic Review. Journal of Functional Morphology and Kinesiology 10(2):133. doi:10.3390/jfmk10020133
3. Lam WWT, Tang YM, Fong KNK (2023). A systematic review of the applications of markerless motion capture (MMC) technology for clinical measurement in rehabilitation. Journal of NeuroEngineering and Rehabilitation 20:57. doi:10.1186/s12984-023-01186-9
4. Schoenwether B, Ripic Z, Nienhuis M, et al. (2025). Reliability of artificial intelligence-driven markerless motion capture in gait analyses of healthy adults. PLOS ONE 20(1):e0316119. doi:10.1371/journal.pone.0316119
5. Alammari B, Schoenwether B, Ripic Z, et al. (2025). Validity of AI-Driven Markerless Motion Capture for Spatiotemporal Gait Analysis in Stroke Survivors. Sensors 25(17):5315. doi:10.3390/s25175315