In recent years, markerless motion capture systems have emerged as a promising alternative to marker-based stereophotogrammetry, enabling motion analysis through monocular video acquisition. However, their biomechanical validity in both controlled and ecological settings remains an open challenge. The aim of this thesis is the study and application of markerless systems in sports contexts, comparing monocular and multi-camera configurations in real-world on-field scenarios. In this context, the present study evaluates the reliability of a monocular 3D pose estimation approach based on SAM3D for the reconstruction of lower-limb kinematics during different motor tasks. First, a gait analysis was conducted on a convenient cohort of healthy subjects and compared with a stereophotogrammetric reference system. Second, side-cut tasks (cutting), typical of rugby and performed on the field, were processed and compared with current state-of-art methods for on-the-field analysis. In both cases, joint kinematics were computed using a consistent biomechanical framework applied to both the reference and markerless data. The results related to gait show good agreement between SAM3D and stereophotogrammetry in the sagittal plane, particularly for hip and knee flexion–extension movements. In contrast, reduced accuracy was observed in the frontal plane, with increased systematic errors and lower correlation. A systematic smoothing effect of the trajectories is also present, resulting in an underestimation of joint range-of-motion peaks, especially during the swing phase. In the rugby tasks, SAM3D exhibited greater variability and lower robustness compared with gait, with performance deterioration potentially associated with high-speed movements, self-occlusions, and unconventional body configurations. Errors are more evident in distal segments and in non-sagittal joint rotations. Overall, SAM3D can provide a coherent representation of the global motor pattern, although its accuracy appears to be strongly dependent on task complexity and environmental conditions.
Negli ultimi anni, i sistemi di motion capture markerless si sono affermati come una promettente alternativa alla stereofotogrammetria marker-based, consentendo l’analisi del movimento tramite acquisizione video monoculare. Tuttavia, la loro validità biomeccanica in contesti sia controllati sia ecologici rimane ancora una sfida aperta. L’obiettivo di questa tesi è lo studio e l’applicazione di sistemi markerless in contesti sportivi, confrontando configurazioni monoculari e multicamera in scenari reali sul campo. In questo contesto, il presente studio valuta l’affidabilità di un approccio di stima della posa 3D monoculare basato su SAM3D per la ricostruzione della cinematica degli arti inferiori in differenti task motori. In primo luogo, è stata condotta un’analisi del cammino su un campione di soggetti sani, confrontata con un sistema di riferimento stereofotogrammetrico. In secondo luogo, sono stati analizzati cambi di direzione, task tipici del rugby, eseguiti on field, e confrontati con le metodologie attualmente considerate stato dell’arte per l’analisi in ambiente sportivo reale. In entrambi i casi, la cinematica articolare è stata calcolata mediante un framework biomeccanico coerente applicato sia ai dati di riferimento sia a quelli markerless. I risultati relativi al cammino evidenziano una buona concordanza tra SAM3D e stereofotogrammetria nel piano sagittale, in particolare per i movimenti di flesso-estensione di anca e ginocchio. Al contrario, si è osservata una ridotta accuratezza nei piani frontale, con un aumento degli errori sistematici e una riduzione della correlazione. È inoltre presente un effetto sistematico di smoothing delle traiettorie, con conseguente sottostima dei picchi di escursione articolare, soprattutto durante la fase di swing. Nei task del rugby, SAM3D ha mostrato una maggiore variabilità e una ridotta robustezza rispetto al cammino, con un peggioramento delle prestazioni potenzialmente associato ad alta velocità, auto-occlusioni e configurazioni corporee non convenzionali. Gli errori risultano più evidenti nei segmenti distali e nelle rotazioni articolari non sagittali. Nel complesso, SAM3D è in grado di fornire una rappresentazione coerente del pattern motorio globale, sebbene l’accuratezza risulti fortemente dipendente dalla complessità del task e dalle condizioni ambientali.
Approccio singola vs multi-camera markerless motion capture nel contesto sportivo: applicazioni sul campo in contesti reali
IMPOCO, CRISTINA
2025/2026
Abstract
In recent years, markerless motion capture systems have emerged as a promising alternative to marker-based stereophotogrammetry, enabling motion analysis through monocular video acquisition. However, their biomechanical validity in both controlled and ecological settings remains an open challenge. The aim of this thesis is the study and application of markerless systems in sports contexts, comparing monocular and multi-camera configurations in real-world on-field scenarios. In this context, the present study evaluates the reliability of a monocular 3D pose estimation approach based on SAM3D for the reconstruction of lower-limb kinematics during different motor tasks. First, a gait analysis was conducted on a convenient cohort of healthy subjects and compared with a stereophotogrammetric reference system. Second, side-cut tasks (cutting), typical of rugby and performed on the field, were processed and compared with current state-of-art methods for on-the-field analysis. In both cases, joint kinematics were computed using a consistent biomechanical framework applied to both the reference and markerless data. The results related to gait show good agreement between SAM3D and stereophotogrammetry in the sagittal plane, particularly for hip and knee flexion–extension movements. In contrast, reduced accuracy was observed in the frontal plane, with increased systematic errors and lower correlation. A systematic smoothing effect of the trajectories is also present, resulting in an underestimation of joint range-of-motion peaks, especially during the swing phase. In the rugby tasks, SAM3D exhibited greater variability and lower robustness compared with gait, with performance deterioration potentially associated with high-speed movements, self-occlusions, and unconventional body configurations. Errors are more evident in distal segments and in non-sagittal joint rotations. Overall, SAM3D can provide a coherent representation of the global motor pattern, although its accuracy appears to be strongly dependent on task complexity and environmental conditions.| File | Dimensione | Formato | |
|---|---|---|---|
|
Impoco_Cristina.pdf
embargo fino al 05/07/2029
Dimensione
3.6 MB
Formato
Adobe PDF
|
3.6 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/109283