Related Experiment Video
Updated: Mar 13, 2026
07:51
Video Movement Analysis Using Smartphones ViMAS: A Pilot Study
Published on: March 14, 2017
17.4K
Evaluation of Smartphone Camera Positioning on Artificial Intelligence Pose Estimation Accuracy for Exercise
Eduarda Oliosi1,2,3, Soraia Ferreira1,4,5, Ana Paula Giordano1,6
1Value for Health CoLAB, 15 Fontes Pereira de Melo Ave, 2nd Fl, Right, Lisbon, 1050‑115, Portugal, 351 937091767.
JMIR Mhealth and Uhealth
|March 11, 2026
Summary
Smartphone camera angle and distance significantly impact AI pose estimation (PE) for exercise tracking. Diagonal and frontal views at mid-range distances (180-200 cm) offer the best accuracy for mobile health apps.
Area of Science:
- Exercise Science
- Computer Vision
- Mobile Health Technology
Background:
- Artificial intelligence (AI)-driven pose estimation (PE) is crucial for scalable exercise tracking in mobile health.
- Occlusion due to camera angle and distance can decrease AI PE accuracy and repetition counting precision.
- The impact of smartphone positioning on AI PE performance requires further controlled investigation.
Purpose of the Study:
- To investigate how smartphone camera angle (front, side, diagonal) and distance affect AI PE detection accuracy and repetition counting precision during push-ups and squats.
- To identify optimal smartphone positioning for reliable exercise tracking in mobile health applications.
Main Methods:
- A within-subject study involving 44 university students performing push-ups or squats.
- Exercises were tracked using AI-based PE across 12 distinct smartphone camera configurations (angles and distances).
- Performance metrics included binary classification accuracy, detection rate, and mean absolute error (MAE) for repetition counting, analyzed using generalized and linear mixed-effects models.
Main Results:
- Overall detection rates were moderate (61.1% for push-ups, 61.5% for squats) with significant mean absolute errors (MAE) in repetition counting.
- Push-ups showed highest accuracy (85.7% detection, MAE=0.28) from diagonal views (90-180 cm) and lowest from front views at 360 cm.
- Squats performed best from diagonal views at 200 cm (95.5% detection, MAE=0.05), with diagonal and front views consistently outperforming side views.
Conclusions:
- Smartphone positioning critically influences the effectiveness of AI-based PE for exercise monitoring.
- Optimal performance for AI exercise tracking is achieved with diagonal and frontal camera views at mid-range distances (180-200 cm).
- Findings provide practical guidance for optimizing mobile health exercise tracking systems.
