Falls in the elderly represent a leading cause of disability, injury, and death. Identification of elderly people at high risk of falling should be a leading medical priority. The postural control system deteriorates with age and disease, balance becomes increasingly tenuous resulting in an enhanced susceptibility to falls. It is important to be able to assess the subject’s ability to maintain balance and to be able to predict the risk of falls. In literature there are many clinical balance tests. Standing Reach (SR) is the motor task analysed in this PhD thesis. The motion is mainly confined to the vertical plane. It consists in leaning the trunk forward trying to reach the maximum displacement of the arms, and maintaining the wrists at approximately the same height during the whole movement. The base of support is fixed during the task because the subject is required not to lift heels or step forward. The maximum displacement of the arms is a measure of clinical significance for its predictive value about recurrent falls in the elderly subjects. Nevertheless, only this measure cannot be considered a measure of dynamic balance, in the sense that it is not able to differentiate healthy elders from individuals with balance impairments. More insight can be obtained from the same motor task if it can be described, in an objective way, looking at its kinematic behaviour too. The first section addresses the compensation problem of skin artifact (ATM). In this section, a compensation technique is proposed based on nonlinear least squares (LSA). The LSA algorithm assumes that the errors due to the ATM can be modeled by an appropriate systematic error and random noise added to the global coordinates of the markers. The technique has been applied to simulated and real data. The results showed the need of more research related to this particular problem of Biomechanics. In the second section, that is mainly related to the falls problem, a markerless approach is illustrated that is able to estimate joint kinematics and the parameters characterizing the SR task; this procedure is based on computer graphics and on computer vision techniques. The method is simple for the end-user and does not require specialized instrumentation. Both quantitative and qualitative characterization of the motor task can be obtained. For the characterization of the SR task the following steps were performed: 1) Video Acquisition. A single webcam was placed orthogonally to the subject’s side and calibrated using the Zangh’s method. 2) Image Segmentation. The frames were segmented by a mixture of K Gaussian distributions (K=35). 3) Body Approximation. The subject was modelled in the sagittal plane as a kinematic chain of 4 segments: foot, leg, trunk, and arm. The body segments have been approximated by ellipses and tracked using a Discrete Curve Evolution algorithm. 4) Parameters estimation. The angles between adjacent segments were calculated using the automatic identification of the major axis of ellipses associated with the relative body segments.The maximum displacement of the arms was determined as Euclidean distance between the final (last frame) and the initial (first frame) position of the distal vertex of the ellipse associated with the arm. Moreover the CoM excursion in the XY-plane and the anterior limit of stability (ASL) has been estimated too. The numerical results and CoM trajectory confirm data found in literature but obtained here with a low cost instrumentation and a completely automatic procedure. The method has been applied in an unstructured environment. In the third section a markerless algorithm, simpler than that described previously, has been proposed for human anthropometric parameters estimation. It is based on the following steps: 1) Image Acquisition. A single digital camera was placed orthogonally to the subject’s side and calibrated using the Zangh’s method. 2) Image segmentation. The acquired images were segmented using Otsu’s method. 3) Body segments approximation. Within each image, strips, each containing a body body segment, have been identified using an algorithm based on gradient. Within each strip the location of the relative body segment was identified using an algorithm based on the cross correlation. Each body segment was then approximated by planar geometric figures. 4) Anthropometric estimation. Once the positions of body segments, and the size of the planar figures that approximate them were determined, the values of the anthropometric parameters were derived. They were compared with those derived by classical models such as those of Winter [106] and De Leva [107], resulting in a more accurate estimation of the proposed algorithm.
Le cadute nei soggetti anziani sono la principale causa di disabilità, malattia e morte. L’identificazione dei soggetti anziani ad alto rischio di caduta riveste quindi un ruolo di primaria importanza. Il sistema di controllo posturale si deteriora con l’età e le malattie, l’equilibrio diminuisce con il conseguente aumento del rischio di caduta. Quindi, è importante avere la capacità di valutare l’abilità del soggetto a mantenere l’equilibrio e di predire il rischio di cadute. In letteratura ci sono molti test clinici per la valutazione dell’equilibrio. Lo Standing Reach (SR) è il task motorio analizzato in questa tesi di dottorato. Il movimento è confinato principalmente nel piano verticale. Lo SR consiste nello spostamento massimale in avanti delle braccia tenute approssimativamente perpendicolari al tronco. Il soggetto non deve alzare i talloni o cambiare base di appoggio. Lo spostamento massimale delle mani è una misura clinicamente significativa per il suo valore predittivo delle cadute di soggetti anziani. Tale misura, di per sé, non può essere considerata una misura dinamica di equilibrio, nel senso che non è in grado di differenziare soggetti anziani sani da quelli con particolari problemi di equilibrio. Maggiori informazioni si possono ottenere dallo stesso task motorio se si riesce a descriverlo valutandone anche gli aspetti cinematici. Il lavoro svolto in questa tesi di dottorato si divide in tre sezioni: Nella prima è stata affrontato il problema della compensazione degli artefatti da tessuto molle (ATM). In tale sezione è stato proposta una tecnica di compensazione basata sui minimi quadrati non lineari (LSA). Nell’LSA si ipotizza che gli errori dovuti agli ATM possano essere modellati mediante opportuno rumore random e sistematico sovrapposto alle coordinate globali dei markers. La tecnica è stata applicata sia ad un caso simulato che a dati reali. I risultati hanno dimostrato la necessità di approfondire la ricerca per quanto riguarda questo particolare problema della Biomeccanica. Nella seconda parte, quella principalmente connessa al problema delle cadute, è stata definita una procedura capace di stimare la cinematica e i parametri caratterizzanti lo SR usando un approccio markerless e adottando le tecniche della computer graphics e della computer vision. Il metodo è semplice e non richiede strumentazione specializzata. Il task motorio è stato caratterizzato sia qualitativamente che quantitativamente. La caratterizzazione dello SR si è basata sulle seguenti fasi: 1) Acquisizione della sequenza video. Una singola webcam è stata posta ortogonalmente al soggetto e calibrata usando il metodo di Zhang. 2) Segmentazione delle immagini. I frames acquisiti sono stati segmentati mediante la mistura di K distribuzioni Gaussiane (K=35). 3) Approssimazione dei segmenti corporei. Il soggetto è stato modellato nel piano sagittale come una catena cinematica costituita da 4 segmenti: piedi, gamba, busto, e braccia. I segmenti corporei sono stati approssimati medianti ellissi, ed il loro tracking è stato eseguito mediante l’algoritmo Evoluzione Discreta di Curve ( DCE). 4) Stima dei parametri. Gli angoli tra segmenti adiacenti sono stati calcolati identificando automaticamente gli assi maggiori delle ellissi associate ai relativi segmenti corporei. Il massimo spostamento delle braccia è stato calcolato come distanza euclidea tra il vertice distale delle ellissi associate al braccio nel primo e nell’ultimo frame. Inoltre, sono stati stimati l’escursione nel piano XY del Centro di Massa (CoM) e il limite di stabilità anteriore (ASL). I risultati numerici e le traiettorie angolari confermano quelli trovati in letteratura, ma in questo caso sono stati ottenuti con una strumentazione low cost e una procedura completamente automatica. Il metodo è stato applicato ad un ambiente non strutturato. Nella terza sezione è stato proposto un algoritmo markerless, semplificato rispetto a quello descritto precedentemente, per la stima dei parametri antropometrici di un soggetto. L’ algoritmo di stima antropometrica è basato sulle seguenti fasi: 1) Acquisizione di immagini. Una singola camera digitale è stata posta ortogonalmente al soggetto e calibrata usando il metodo di Zhang. 2) Segmentazione delle immagini. Le immagini acquisite sono state segmentate mediante il metodo di Otsu. 3) Approssimazione dei segmenti corporei. All’interno di ogni immagine, sono state individuate delle fasce contenenti ciascuna dei segmenti corporei, mediante un algoritmo basato sul gradiente. All’interno di ciascun fascia è stata individuata la posizione del relativo segmento corporeo, mediante un algoritmo basato sulla cross correlazione. Ogni segmento corporeo è stato poi approssimato mediante figure geometriche piane. 4) Stima antropometrica. Note le posizioni dei segmenti corporei, e le dimensioni delle figure piane che li approssima, sono stati ricavati i parametri antropometrici desiderati. I parametri antropometrici stimati sono stati confrontati con quelli ricavati dai modelli classici quali quelli di Winter [106] e De Leva [107] dimostrando cosi la maggior precisione dell’algoritmo di stima proposto.
Analisi del movimento umano mediante approccio Markerless(2011 Jan 19).
Analisi del movimento umano mediante approccio Markerless
-
2011-01-19
Abstract
Falls in the elderly represent a leading cause of disability, injury, and death. Identification of elderly people at high risk of falling should be a leading medical priority. The postural control system deteriorates with age and disease, balance becomes increasingly tenuous resulting in an enhanced susceptibility to falls. It is important to be able to assess the subject’s ability to maintain balance and to be able to predict the risk of falls. In literature there are many clinical balance tests. Standing Reach (SR) is the motor task analysed in this PhD thesis. The motion is mainly confined to the vertical plane. It consists in leaning the trunk forward trying to reach the maximum displacement of the arms, and maintaining the wrists at approximately the same height during the whole movement. The base of support is fixed during the task because the subject is required not to lift heels or step forward. The maximum displacement of the arms is a measure of clinical significance for its predictive value about recurrent falls in the elderly subjects. Nevertheless, only this measure cannot be considered a measure of dynamic balance, in the sense that it is not able to differentiate healthy elders from individuals with balance impairments. More insight can be obtained from the same motor task if it can be described, in an objective way, looking at its kinematic behaviour too. The first section addresses the compensation problem of skin artifact (ATM). In this section, a compensation technique is proposed based on nonlinear least squares (LSA). The LSA algorithm assumes that the errors due to the ATM can be modeled by an appropriate systematic error and random noise added to the global coordinates of the markers. The technique has been applied to simulated and real data. The results showed the need of more research related to this particular problem of Biomechanics. In the second section, that is mainly related to the falls problem, a markerless approach is illustrated that is able to estimate joint kinematics and the parameters characterizing the SR task; this procedure is based on computer graphics and on computer vision techniques. The method is simple for the end-user and does not require specialized instrumentation. Both quantitative and qualitative characterization of the motor task can be obtained. For the characterization of the SR task the following steps were performed: 1) Video Acquisition. A single webcam was placed orthogonally to the subject’s side and calibrated using the Zangh’s method. 2) Image Segmentation. The frames were segmented by a mixture of K Gaussian distributions (K=35). 3) Body Approximation. The subject was modelled in the sagittal plane as a kinematic chain of 4 segments: foot, leg, trunk, and arm. The body segments have been approximated by ellipses and tracked using a Discrete Curve Evolution algorithm. 4) Parameters estimation. The angles between adjacent segments were calculated using the automatic identification of the major axis of ellipses associated with the relative body segments.The maximum displacement of the arms was determined as Euclidean distance between the final (last frame) and the initial (first frame) position of the distal vertex of the ellipse associated with the arm. Moreover the CoM excursion in the XY-plane and the anterior limit of stability (ASL) has been estimated too. The numerical results and CoM trajectory confirm data found in literature but obtained here with a low cost instrumentation and a completely automatic procedure. The method has been applied in an unstructured environment. In the third section a markerless algorithm, simpler than that described previously, has been proposed for human anthropometric parameters estimation. It is based on the following steps: 1) Image Acquisition. A single digital camera was placed orthogonally to the subject’s side and calibrated using the Zangh’s method. 2) Image segmentation. The acquired images were segmented using Otsu’s method. 3) Body segments approximation. Within each image, strips, each containing a body body segment, have been identified using an algorithm based on gradient. Within each strip the location of the relative body segment was identified using an algorithm based on the cross correlation. Each body segment was then approximated by planar geometric figures. 4) Anthropometric estimation. Once the positions of body segments, and the size of the planar figures that approximate them were determined, the values of the anthropometric parameters were derived. They were compared with those derived by classical models such as those of Winter [106] and De Leva [107], resulting in a more accurate estimation of the proposed algorithm.File | Dimensione | Formato | |
---|---|---|---|
Tesi.Sgattoni.pdf
Solo gestori archivio
Tipologia:
Tesi di dottorato
Licenza d'uso:
Non specificato
Dimensione
3.06 MB
Formato
Adobe PDF
|
3.06 MB | Adobe PDF | Visualizza/Apri Richiedi una copia |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.