AI Rep Counting for Home Exercise Programs: How It Works and What the Evidence Says

Key points

• AI rep counting uses pose estimation, a form of computer vision, to follow the body in phone-camera video and count each repetition of a home exercise automatically.
• It addresses a real blind spot: non-adherence to home-based physical therapy can reach 70%, and in one study exercise diaries overstated what patients did compared with concealed sensors.
• The accuracy research is early and setup-dependent. In one smartphone study, camera position strongly affected how reliably reps were detected.
• Measured completion supports the conversation about adherence; it doesn't replace it.

The home exercise blind spot

Most of a rehabilitation program happens between visits, in the exercises a patient does at home. A systematic review of home-based physical therapies found that non-adherence can reach 70%; the strongest predictors of adherence were intention, self-motivation, self-efficacy, previous adherence to exercise and social support (Essery et al., 2017).

The harder problem is that clinicians usually can't see what happened. Adherence is typically recorded by self-report: a log, a diary or an answer to "how have the exercises been going?" In 54 adults with chronic knee pain, Nicolson and colleagues hid accelerometers in exercise equipment and compared them with patients' diaries over 12 weeks. The diaries overestimated adherence: a median of 220 exercises recorded versus 176 measured. The two only moderately agreed. Self-report isn't useless, but it isn't a measurement. We cover the wider adherence problem in closing the home exercise adherence gap.

How AI rep counting works

1. Capture. The patient props up a phone so the camera can see them, and starts the assigned exercise.
2. Pose estimation. A computer-vision model finds body landmarks (shoulders, hips, knees, ankles and so on) in every video frame. No markers or wearable sensors are needed; it's the same markerless approach used for AI movement analysis.
3. Repetition detection. As a joint angle or body position cycles through the movement, each completed cycle is counted as a repetition.
4. Record. Reps, sets and sessions are logged, so the clinician sees what was completed rather than what was remembered.

How accurate is it?

The honest answer is that the evidence is early and depends on the setup:

• Camera position matters. In 44 healthy university students doing push-ups and squats, a study testing 12 smartphone positions found mean detection rates of about 61% across positions, and concluded that accuracy is significantly affected by where the phone is placed (Oliosi et al., 2026). Guided setup, with the camera placed where the movement is visible, isn't a detail.
• Older adults. In seven care-home residents aged 65–94 doing six rehabilitation exercises, a single-camera deep-learning approach correlated with sensor-based ground truth at 0.86 for repetition count (Stefanova et al., 2026). The sample was very small.

Most published studies are small, and many use healthy volunteers or fitness exercises. Clinic-ready rep counting still has to show it works for the exercises and patients it's used with.

Does tracking improve adherence?

Here the evidence is mixed, and it's worth being clear about that:

• A systematic review of 10 randomized trials (1,117 participants) found that 7 reported a significant difference in adherence, and concluded that digital add-ons can likely increase home-exercise adherence in the short term, with longer-term effects less certain (Lang et al., 2022).
• A review focused on chronic musculoskeletal pain found no evidence that digital interventions enhance adherence to therapeutic exercise (Bunting et al., 2021).
• A review of mobile apps for musculoskeletal pain found they may help reduce pain and improve function, but evidence on adherence was sparse (Thompson et al., 2023).

The reasonable reading: measurement makes adherence visible. What changes behavior is what the clinician and patient do with it, such as adjusting a program that isn't landing, or recognizing progress.

Where it fits in a plan of care

• Between visits: see completed reps and sessions before the next appointment, not at the re-evaluation.
• In the conversation: "you did about half the prescribed sets this week" is a better starting point than "have you been doing your exercises?"
• Remote therapeutic monitoring: the CMS descriptor for musculoskeletal RTM device supply (CPT 98977) lists therapy adherence as an example of what RTM monitors, and RTM must be furnished under a therapy plan of care. Billing decisions remain with the practice; see our RTM CPT code reference and RTM guide.

How Kinetically does it

In Kinetically, patients do their assigned home exercise program in the app and the phone camera counts repetitions automatically, so the therapist sees completed reps, sessions and adherence alongside short home re-tests of the measures that matter, such as gait speed and sit-to-stand. See home programs and treatments or Kinetically for physical therapy.

Rep counting is a measurement tool that supports clinical decisions; it is not a substitute for the therapist's judgment about the program or the patient.

References

1. Essery R, Geraghty AWA, Kirby S, Yardley L. Predictors of adherence to home-based physical therapies: a systematic review. Disabil Rehabil. 2017;39(6):519-534. doi:10.3109/09638288.2016.1153160
2. Nicolson PJA, Hinman RS, Wrigley TV, Stratford PW, Bennell KL. Self-reported home exercise adherence: a validity and reliability study using concealed accelerometers. J Orthop Sports Phys Ther. 2018;48(12):943-950. doi:10.2519/jospt.2018.8275
3. Oliosi E, Ferreira S, Giordano AP, et al. Evaluation of smartphone camera positioning on artificial intelligence pose estimation accuracy for exercise detection: observational study. JMIR mHealth uHealth. 2026;14:e82412. doi:10.2196/82412
4. Stefanova V, Maeyens E, Brughmans J, et al. Deep learning-based physical exercise assessment of older adults using single-camera videos. PLOS Digit Health. 2026;5(8):e0001576. doi:10.1371/journal.pdig.0001576
5. Lang S, McLelland C, MacDonald D, Hamilton DF. Do digital interventions increase adherence to home exercise rehabilitation? A systematic review of randomised controlled trials. Arch Physiother. 2022;12:24. doi:10.1186/s40945-022-00148-z
6. Bunting JW, Withers TM, Heneghan NR, Greaves CJ. Digital interventions for promoting exercise adherence in chronic musculoskeletal pain: a systematic review and meta-analysis. Physiotherapy. 2021;111:23-30. doi:10.1016/j.physio.2020.08.001
7. Thompson D, Rattu S, Tower J, Egerton T, et al. Mobile app use to support therapeutic exercise for musculoskeletal pain conditions may help improve pain intensity and self-reported physical function: a systematic review. J Physiother. 2023;69(1):23-34. doi:10.1016/j.jphys.2022.11.012
8. Centers for Medicare & Medicaid Services. MLN Matters MM14250: Therapy Code List, 2026 Annual Update. cms.gov