In recent years, motion analysis has gained increasing relevance across various disciplines, particularly in clinical and sports fields. To date, traditional systems rely on optoelectronic technologies that require the use of markers. Despite their excellent performance, the need for dedicated laboratories and the placement of markers on the subject’s body have driven the development of new technologies capable of overcoming these limitations. In this context, inertial measurement units (IMUs) and video-based markerless systems have emerged. This study aims to validate the markerless motion analysis software CapturyStudio (The Captury GmbH, Saarbrüken, Germany), which is based on a multi-camera video setup for reconstructing the joint kinematics of both the upper and lower limbs. The software operates on a passive vision system that employs visual hull and background subtraction methods to extract the subject’s silhouette. This silhouette is then used to automatically scale and adapt Captury’s template skeleton to the subject under analysis. The scaling process involves estimating joint center positions using multiple 3D Gaussian functions and local optimization procedures. In this study, video data were acquired using 8 BTS SMART EVO-DX 2 cameras (BTS Bioengineering S.p.A., Garbagnate Milanese (MI), Italy), which allow both marker-based and markerless acquisitions. To validate the system’s performance, the obtained data were compared with those acquired through the Xsens inertial sensor system (Movella, Henderson (NV), USA). The study was conducted at the Human Performance Lab of Politecnico di Milano and involved a total of 10 healthy subjects. The acquisition protocol included two different exercises commonly used in clinical and rehabilitation settings: one for the upper limb and one for the lower limb. The investigated joint angles were flexion-extension angles of the elbow, shoulder, knee, and hip. CapturyStudio demonstrated good performance in tracking the movements of both the upper and lower limbs, as evidenced by root mean square error values, both in absolute and percentage terms (RMSE and %RMSE), as well as percentage accuracy values (Accuracy) estimated from the range of motion (ROM) of the analyzed joints. Additionally, intraclass correlation coefficients (ICC) and Spearman correlation values suggest high reliability of the markerless signal and a strong correlation between the two acquired signals, with better results observed for the upper limb. Non-parametric statistical tests indicated good equivalence in system performance between the right and left sides of the body, while highlighting a significant difference in ROM estimation between Xsens and CapturyStudio for the lower limb. This study establishes the basis for future research aimed at evaluating the system’s performance in estimating joint angles in the frontal and transverse planes, under different environmental conditions, or in pathological subjects.
Nel corso degli ultimi anni l’analisi del movimento ha acquisito sempre maggiore rilevanza in diversi ambiti disciplinari, in particolare negli ambiti clinico e sportivo. Ad oggi le tecnologie tradizionali si basano su sistemi optoelettronici che prevedono l’utilizzo di marcatori. Nonostante le ottime prestazioni dimostrate, la necessità di laboratori specifici e il posizionamento di marcatori sul corpo del soggetto hanno portato allo sviluppo di nuove tecnologie, capaci di superare tali limitazioni. In questo contesto si inseriscono le unità di misura inerziali (IMU), e i sistemi markerless basati su video. Questo studio si pone l’obiettivo di validare il software di analisi del movimento markerless CapturyStudio (The Captury GmbH, Saarbrüken, Germania) basato su video multi-camera, per la ricostruzione della cinematica articolare sia degli arti superiori che degli arti inferiori. Il software si basa su un sistema di visione passiva, che utilizza metodi di inviluppo visivo e sottrazione dello sfondo per identificare la silhouette del soggetto. Questa viene sfruttata per scalare in maniera automatica il modello scheletrico specifico di Captury, e adattarlo al soggetto in esame. Il processo di scalatura implica la stima delle posizioni dei diversi centri articolari mediante funzioni di Gauss 3D multiple e procedure di ottimizzazione locale. Nello studio, i video sono stati acquisiti attraverso 8 telecamere BTS SMART EVO-DX 2 (BTS Bioengineering S.p.A., Garbagnate Milanese (MI), Italia) per consentire acquisizioni sia marker-based che markerless. Per validare le prestazioni del sistema, i dati ottenuti sono stati confrontati con i dati ottenuti tramite il sistema di sensori inerziali sviluppato da Xsens (Movella, Henderson (NV), USA). Lo studio è stato condotto presso lo Human Performance Lab del Politecnico di Milano, e ha coinvolto un totale di 10 soggetti sani. Il protocollo seguito durante le sessioni di acquisizione ha previsto l’esecuzione di due esercizi differenti comunemente svolti in ambito clinico-riabilitativo, uno per l’analisi degli arti superiori e uno per gli arti inferiori. Gli angoli articolari indagati sono stati rispettivamente gli angoli di flessione-estensione del gomito, della spalla, del ginocchio e dell’anca. Il software CapturyStudio ha dimostrato buone prestazioni nel tracciamento dei movimenti sia degli arti superiori, sia degli arti inferiori, come dimostrato dai valori di errore quadratico medio sia in termini assoluti, che percentuali (RMSE e %RMSE) e dai valori di accuratezza percentuale (Accuracy) stimati a partire dai range of motion (ROM) delle diverse articolazioni. I valori relativi ai coefficienti di correlazione intraclasse (ICC) e di Spearman suggeriscono inoltre un’ottima affidabilità del segnale markerless, e una forte correlazione tra i due segnali acquisiti, sebbene i risultati appaiano migliori per l’arto superiore. Test statistici non parametrici hanno dimostrato in generale una buona equivalenza delle prestazioni del sistema tra lato destro e sinistro del corpo, mentre hanno evidenziato una differenza significativa nella stima del ROM articolare tra i sistemi Xsens e CapturyStudio, relativamente all’arto inferiore. Tale lavoro pone le basi per ricerche future, volte a valutare le prestazioni del sistema nella stima degli angoli articolari eseguiti nel piano frontale e trasversale, in condizioni ambientali differenti, o su soggetti patologici.
Markerless motion capture: multi-camera system validation for upper and lower limb assessment
Strada, Gaia
2023/2024
Abstract
In recent years, motion analysis has gained increasing relevance across various disciplines, particularly in clinical and sports fields. To date, traditional systems rely on optoelectronic technologies that require the use of markers. Despite their excellent performance, the need for dedicated laboratories and the placement of markers on the subject’s body have driven the development of new technologies capable of overcoming these limitations. In this context, inertial measurement units (IMUs) and video-based markerless systems have emerged. This study aims to validate the markerless motion analysis software CapturyStudio (The Captury GmbH, Saarbrüken, Germany), which is based on a multi-camera video setup for reconstructing the joint kinematics of both the upper and lower limbs. The software operates on a passive vision system that employs visual hull and background subtraction methods to extract the subject’s silhouette. This silhouette is then used to automatically scale and adapt Captury’s template skeleton to the subject under analysis. The scaling process involves estimating joint center positions using multiple 3D Gaussian functions and local optimization procedures. In this study, video data were acquired using 8 BTS SMART EVO-DX 2 cameras (BTS Bioengineering S.p.A., Garbagnate Milanese (MI), Italy), which allow both marker-based and markerless acquisitions. To validate the system’s performance, the obtained data were compared with those acquired through the Xsens inertial sensor system (Movella, Henderson (NV), USA). The study was conducted at the Human Performance Lab of Politecnico di Milano and involved a total of 10 healthy subjects. The acquisition protocol included two different exercises commonly used in clinical and rehabilitation settings: one for the upper limb and one for the lower limb. The investigated joint angles were flexion-extension angles of the elbow, shoulder, knee, and hip. CapturyStudio demonstrated good performance in tracking the movements of both the upper and lower limbs, as evidenced by root mean square error values, both in absolute and percentage terms (RMSE and %RMSE), as well as percentage accuracy values (Accuracy) estimated from the range of motion (ROM) of the analyzed joints. Additionally, intraclass correlation coefficients (ICC) and Spearman correlation values suggest high reliability of the markerless signal and a strong correlation between the two acquired signals, with better results observed for the upper limb. Non-parametric statistical tests indicated good equivalence in system performance between the right and left sides of the body, while highlighting a significant difference in ROM estimation between Xsens and CapturyStudio for the lower limb. This study establishes the basis for future research aimed at evaluating the system’s performance in estimating joint angles in the frontal and transverse planes, under different environmental conditions, or in pathological subjects.File | Dimensione | Formato | |
---|---|---|---|
2025_04_Strada_Tesi.pdf
accessibile in internet per tutti a partire dal 04/03/2026
Descrizione: Testo tesi
Dimensione
5.58 MB
Formato
Adobe PDF
|
5.58 MB | Adobe PDF | Visualizza/Apri |
2025_04_Strada_Executive_Summary.pdf
accessibile in internet per tutti a partire dal 04/03/2026
Descrizione: Testo executive summary
Dimensione
765.43 kB
Formato
Adobe PDF
|
765.43 kB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/235894