Validation of video-based event detection against laboratory-standard biomechanical measures: a synchronized multimodal study
5 affiliations
- 1Kunming Medical University Affiliated Qujing Hospital, Qujing, China
- 2The Second Affiliated Hospital, University of South China, Hengyang, China
- 3Department of Orthopedics, Qujing Central Hospital of Yunnan Province, Qujing, China
- 4Chinese PLA General Hospital, Beijing, China
- 5Department of Orthopedics, The First Affiliated Hospital of USTC, Division of Life Sciences and Medicine, University of Science and Technology of China, Hefei, China
Abstract
IntroductionBiomechanical event detection is fundamental to human movement analysis in clinical, rehabilitation, and sports settings, yet conventional laboratory systems remain costly and inaccessible in many contexts. This study validated a markerless, video-primary event detection system against laboratory-reference biomechanical measurements using synchronized multimodal data.MethodsEight healthy adults performed standardized upper-limb dumbbell lifting tasks in a controlled laboratory while video, Vicon motion-capture, and surface electromyography (EMG) data were recorded (76 valid trials; 304 event labels). A hybrid pipeline combining threshold-based peak detection with Long Short-Term Memory (LSTM) networks identified four events: action start, peak of rising, end of holding, and start of returning. Kinematic features were derived from video; two EMG envelope channels were additionally used in the LSTM refinement stage. Streams were synchronized using hardware triggers (Vicon and EMG), software-started video recording, and cross-correlation of a shared kinematic signal prior to independent event detection in each system. Video-based event timestamps were compared with Vicon reference timestamps without further reference-informed adjustment, using intraclass correlation coefficients (ICC), Bland–Altman analysis, mean absolute error (MAE), root mean square error (RMSE), and classification metrics.ResultsThe video-based system showed excellent agreement with the Vicon reference (ICC: 0.90–0.94), low temporal error (MAE: 38–51 ms; RMSE: 52–70 ms), and minimal systematic bias; errors corresponded to approximately 1.1–1.5 frames at the 30 fps acquisition rate. Classification performance was strong across all events (F1 score: 93.4%; sensitivity: 93.4%; specificity: 97.8%). EMG activation peaks correlated with detected events (r = 0.86–0.91); because EMG envelopes also served as LSTM inputs, this is an internal consistency check rather than independent physiological validation.DiscussionThese findings indicate that a hybrid video-primary approach can achieve accuracy comparable to laboratory-reference measurements for upper-limb analysis under controlled conditions, offering a scalable complement to traditional motion-capture systems. An ablation of the EMG input channels to establish video-only performance, and validation in larger, more diverse populations and real-world settings, are required.
Related
- Multi-scale lodging in rice and wheat: bridging biomechanical traits, phenotypic detection, and intelligent harvesting systems
- The limit of detection of a Phytophthora agathidicida LAMP assay is higher under alternative laboratory conditions.
- Cost-efficient filtration and replication strategies match or exceed detection sensitivity for marine eDNA diversity at an intertidal site under a standardized laboratory workflow
- Validation of EUCAST 2026 breakpoints for Brucella species using routine laboratory methods: A comprehensive French National Reference Centre study
- The Canadian Pediatric Imaging Platform (C-PIP) Reference Sample: Neuroimaging Protocol and Cross-site Validation for Harmonized Multimodal Pediatric Imaging
BibTeX
@article{xiao2026validation,
title = {{Validation of video-based event detection against laboratory-standard biomechanical measures: a synchronized multimodal study}},
author = {Zhenping Xiao and Zhenping Xiao and Zhenping Xiao and Zaitian Huang and Henrui Zhang and Henrui Zhang and Henrui Zhang and Daiyuan Tang and Daiyuan Tang and Jingyu Li and Yangjie Li and Chengkai He and Li Tan and Fei He and Fei He},
journal = {Frontiers in Bioengineering and Biotechnology},
year = {2026},
month = oct,
doi = {10.3389/fbioe.2026.1909090},
url = {https://doi.org/10.3389/fbioe.2026.1909090},
note = {CC BY},
}
RIS
TY - JOUR
TI - Validation of video-based event detection against laboratory-standard biomechanical measures: a synchronized multimodal study
AU - Zhenping Xiao
AU - Zhenping Xiao
AU - Zhenping Xiao
AU - Zaitian Huang
AU - Henrui Zhang
AU - Henrui Zhang
AU - Henrui Zhang
AU - Daiyuan Tang
AU - Daiyuan Tang
AU - Jingyu Li
AU - Yangjie Li
AU - Chengkai He
AU - Li Tan
AU - Fei He
AU - Fei He
PY - 2026
DA - 2026/10/26
JO - Frontiers in Bioengineering and Biotechnology
DO - 10.3389/fbioe.2026.1909090
UR - https://doi.org/10.3389/fbioe.2026.1909090
AB - IntroductionBiomechanical event detection is fundamental to human movement analysis in clinical, rehabilitation, and sports settings, yet conventional laboratory systems remain costly and inaccessible in many contexts. This study validated a markerless, video-primary event detection system against laboratory-reference biomechanical measurements using synchronized multimodal data.MethodsEight healthy adults performed standardized upper-limb dumbbell lifting tasks in a controlled laboratory while video, Vicon motion-capture, and surface electromyography (EMG) data were recorded (76 valid trials; 304 event labels). A hybrid pipeline combining threshold-based peak detection with Long Short-Term Memory (LSTM) networks identified four events: action start, peak of rising, end of holding, and start of returning. Kinematic features were derived from video; two EMG envelope channels were additionally used in the LSTM refinement stage. Streams were synchronized using hardware triggers (Vicon and EMG), software-started video recording, and cross-correlation of a shared kinematic signal prior to independent event detection in each system. Video-based event timestamps were compared with Vicon reference timestamps without further reference-informed adjustment, using intraclass correlation coefficients (ICC), Bland–Altman analysis, mean absolute error (MAE), root mean square error (RMSE), and classification metrics.ResultsThe video-based system showed excellent agreement with the Vicon reference (ICC: 0.90–0.94), low temporal error (MAE: 38–51 ms; RMSE: 52–70 ms), and minimal systematic bias; errors corresponded to approximately 1.1–1.5 frames at the 30 fps acquisition rate. Classification performance was strong across all events (F1 score: 93.4%; sensitivity: 93.4%; specificity: 97.8%). EMG activation peaks correlated with detected events (r = 0.86–0.91); because EMG envelopes also served as LSTM inputs, this is an internal consistency check rather than independent physiological validation.DiscussionThese findings indicate that a hybrid video-primary approach can achieve accuracy comparable to laboratory-reference measurements for upper-limb analysis under controlled conditions, offering a scalable complement to traditional motion-capture systems. An ablation of the EMG input channels to establish video-only performance, and validation in larger, more diverse populations and real-world settings, are required.
SN - 2296-4185
ER -