Publicação
System for automated motion analysis during gait using an RGB-D camera
| Resumo: | Nowadays it is still common in clinical practice to assess the gait (or way of walking) of a given subject through the visual observation and use of a rating scale, which is a subjective approach. However, sensors including RGB-D cameras, such as the Microsoft Kinect, can be used to obtain quantitative information that allows performing gait analysis in a more objective way. The quantitative gait analysis results can be very useful for example to support the clinical assessment of patients with diseases that can affect their gait, such as Parkinson’s disease. The main motivation of this thesis was thus to provide support to gait assessment, by allowing to carry out quantitative gait analysis in an automated way. This objective was achieved by using 3-D data, provided by a single RGB-D camera, to automatically select the data corresponding to walking and then detect the gait cycles performed by the subject while walking. For each detected gait cycle, we obtain several gait parameters, which are used together with anthropometric measures to automatically identify the subject being assessed. The automated gait data selection relies on machine learning techniques to recognize three different activities (walking, standing, and marching), as well as two different positions of the subject in relation to the camera (facing the camera and facing away from it). For gait cycle detection, we developed an algorithm that estimates the instants corresponding to given gait events. The subject identification based on gait is enabled by a solution that was also developed by relying on machine learning. The developed solutions were integrated into a system for automated gait analysis, which we found to be a viable alternative to gold standard systems for obtaining several spatiotemporal and some kinematic gait parameters. Furthermore, the system is suitable for use in clinical environments, as well as ambulatory scenarios, since it relies on a single markerless RGB-D camera that is less expensive, more portable, less intrusive and easier to set up, when compared with the gold standard systems (multiple cameras and several markers attached to the subject’s body). |
|---|---|
| Autores principais: | Rocha, Ana Patrícia Oliveira Ferreira da |
| Assunto: | 3-D human motion analysis Automated gait analysis RGB-D camera Machine learning Activity recognition Human identification |
| Ano: | 2019 |
| País: | Portugal |
| Tipo de documento: | tese de doutoramento |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade de Aveiro |
| Idioma: | inglês |
| Origem: | RIA - Repositório Institucional da Universidade de Aveiro |
| Resumo: | Nowadays it is still common in clinical practice to assess the gait (or way of walking) of a given subject through the visual observation and use of a rating scale, which is a subjective approach. However, sensors including RGB-D cameras, such as the Microsoft Kinect, can be used to obtain quantitative information that allows performing gait analysis in a more objective way. The quantitative gait analysis results can be very useful for example to support the clinical assessment of patients with diseases that can affect their gait, such as Parkinson’s disease. The main motivation of this thesis was thus to provide support to gait assessment, by allowing to carry out quantitative gait analysis in an automated way. This objective was achieved by using 3-D data, provided by a single RGB-D camera, to automatically select the data corresponding to walking and then detect the gait cycles performed by the subject while walking. For each detected gait cycle, we obtain several gait parameters, which are used together with anthropometric measures to automatically identify the subject being assessed. The automated gait data selection relies on machine learning techniques to recognize three different activities (walking, standing, and marching), as well as two different positions of the subject in relation to the camera (facing the camera and facing away from it). For gait cycle detection, we developed an algorithm that estimates the instants corresponding to given gait events. The subject identification based on gait is enabled by a solution that was also developed by relying on machine learning. The developed solutions were integrated into a system for automated gait analysis, which we found to be a viable alternative to gold standard systems for obtaining several spatiotemporal and some kinematic gait parameters. Furthermore, the system is suitable for use in clinical environments, as well as ambulatory scenarios, since it relies on a single markerless RGB-D camera that is less expensive, more portable, less intrusive and easier to set up, when compared with the gold standard systems (multiple cameras and several markers attached to the subject’s body). |
|---|