What question did this study set out to answer?

This research aims to assess the accuracy of markerless motion capture for measuring upper limb kinematics in hemiplegic stroke survivors.

May 8, 2026

Abstract Number: Esoc2026a1109 the Validity of Markerless Motion Capture for Quantifying Upper Limb Kinematics in Hemiplegic Stroke Survivors

Key Points

This research aims to assess the accuracy of markerless motion capture for measuring upper limb kinematics in hemiplegic stroke survivors.
Thirteen stroke survivors (7 male, 5 female) performed a drinking task while their movements were recorded using Vicon and markerless motion capture.
Participants were grouped by Fugl–Meyer Upper Limb score ranging from mild to severe impairments.
Agreement between kinematics measures was evaluated using RMSE, Concordance Correlation Coefficient, and Pearson’s r.
Shoulder and elbow flexion measurements showed high similarity (r=0.90–0.99) and acceptable RMSE values (shoulder 6.4–12.5°, elbow 11.7–13.4°).
Normalised wrist tangential velocity had the best agreement with RMSE 0.1–0.3 and CCC 0.89–0.97.
Wrist flexion had the poorest agreement, especially in moderate/severe stroke groups (RMSE 11.6–20.1°, CCC 0.14–0.63).

Abstract

Abstract Background and aims Arm weakness is common after stroke. Movement kinematics could identify intervention targets but are currently unavailable to clinicians. We evaluated accessible video-based markerless motion capture (MMC) accuracy for upper-limb kinematics in hemiplegic stroke survivors. Methods Thirteen stroke survivors (male/female: 7/5, age: 62.3 (6.2) years, recruitment ongoing) performed a drinking task. Eleven were included in the affected arm analysis and were grouped by Fugl–Meyer Upper Limb score: mild (51–60, n=2), moderate (26–50, n=5), and severe (0–25, n=4). Motion data was recorded concurrently using Vicon (120 Hz) and MMC, using two RealSense RGB-depth cameras (30 Hz) and RTMPose. Shoulder, elbow, and wrist angles and normalised wrist tangential velocity (TVW) were computed for affected and unaffected arms. Agreement was assessed using RMSE, Concordance Correlation Coefficient (CCC), and Pearson’s r. Results Fig 1 presents group-based results. Across groups, shoulder and elbow flexion showed strong similarity (r=0.90–0.99) with moderate RMSE (shoulder 6.4–12.5°, elbow 11.7–13.4°) and variable CCC (shoulder 0.52–0.94; elbow 0.08–0.96). TVW showed the best agreement (RMSE 0.1–0.3; CCC 0.89–0.97; r 0.87–0.96). Wrist flexion showed the poorest agreement, especially in moderate/severe stroke (RMSE 11.6–20.1°; CCC 0.14–0.63; r 0.24–0.82). Shoulder abduction varied in relation to severity (RMSE 2.8–6.5°; CCC 0.47–0.90; r 0.72–0.95). Conclusions Accessible MMC techniques show promise for quantifying post-stroke upper-limb movement, but agreement varies between joints and declines with impairment. Performance could be improved through alternative camera setups and optimised MMC approaches. Conflict of interest Rylea Hart: Funding provided by Private Physiotherapy Educational Fund (PPEF) Scheme C Award 2024 and the Physiotherapy Research Fund Award PRF /24A/POD01. No further disclosures. Sera Boston: Funding provided by Private Physiotherapy Educational Fund (PPEF) Scheme C Award 2024. No further disclosures. Letizia Gionfrida: Nothing to disclose. Irene Di Giulio: Nothing to disclose. Ulrike Hammerbeck: Funding provided by Private Physiotherapy Educational Fund (PPEF) Scheme C Award 2024 and the Physiotherapy Research Fund Award PRF /24A/POD01. No further disclosures. Figure 1 - belongs to Results

Bookmark

Abstract Number: Esoc2026a1109 the Validity of Markerless Motion Capture for Quantifying Upper Limb Kinematics in Hemiplegic Stroke Survivors

Key Points

Abstract

Cite This Study

Also Consider

Also Consider