Measuring and understanding human motion is crucial in several domains, ranging from neuroscience, to rehabilitation and sports biomechanics. Quantitative information about human motion is fundamental to study how our Central Nervous System controls and organizes movements to functionally evaluate motor performance and deficits. In the last decades, the research in this field has made considerable progress. State-of-the-art technologies that provide useful and accurate quantitative measures rely on marker-based systems. Unfortunately, markers are intrusive and their number and location must be determined a priori. Also, marker-based systems require expensive laboratory settings with several infrared cameras. This could modify the naturalness of a subject’s movements and induce discomfort. Last, but not less important, they are computationally expensive in time and space. Recent advances on markerless pose estimation based on computer vision and deep neural networks are opening the possibility of adopting efficient video-based methods for extracting movement information from RGB video data. In this contest, this thesis presents original contributions to the following objectives: (i) the implementation of a video-based markerless pipeline to quantitatively characterize human motion; (ii) the assessment of its accuracy if compared with a gold standard marker-based system; (iii) the application of the pipeline to different domains in order to verify its versatility, with a special focus on the characterization of the motion of preterm infants and on gait analysis. With the proposed approach we highlight that, starting only from RGB videos and leveraging computer vision and machine learning techniques, it is possible to extract reliable information characterizing human motion comparable to that obtained with gold standard marker-based systems.
Markerless Human Motion Analysis
MORO, MATTEO
2022
Abstract
Measuring and understanding human motion is crucial in several domains, ranging from neuroscience, to rehabilitation and sports biomechanics. Quantitative information about human motion is fundamental to study how our Central Nervous System controls and organizes movements to functionally evaluate motor performance and deficits. In the last decades, the research in this field has made considerable progress. State-of-the-art technologies that provide useful and accurate quantitative measures rely on marker-based systems. Unfortunately, markers are intrusive and their number and location must be determined a priori. Also, marker-based systems require expensive laboratory settings with several infrared cameras. This could modify the naturalness of a subject’s movements and induce discomfort. Last, but not less important, they are computationally expensive in time and space. Recent advances on markerless pose estimation based on computer vision and deep neural networks are opening the possibility of adopting efficient video-based methods for extracting movement information from RGB video data. In this contest, this thesis presents original contributions to the following objectives: (i) the implementation of a video-based markerless pipeline to quantitatively characterize human motion; (ii) the assessment of its accuracy if compared with a gold standard marker-based system; (iii) the application of the pipeline to different domains in order to verify its versatility, with a special focus on the characterization of the motion of preterm infants and on gait analysis. With the proposed approach we highlight that, starting only from RGB videos and leveraging computer vision and machine learning techniques, it is possible to extract reliable information characterizing human motion comparable to that obtained with gold standard marker-based systems.File | Dimensione | Formato | |
---|---|---|---|
phdunige_3903881.pdf
accesso aperto
Dimensione
2.36 MB
Formato
Adobe PDF
|
2.36 MB | Adobe PDF | Visualizza/Apri |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14242/70986
URN:NBN:IT:UNIGE-70986