| Resumo: | Gait has been an extensively investigated topic in recent years. Through the analysis of gait it is possible to detect pathologies, which makes this analysis very important to assess anomalies and, consequently, help in the diagnosis and rehabilitation of patients. There are some systems for analyzing gait, but they are usually either systems with subjective evaluations or systems used in specialized laboratories with complex equipment, which makes them very expensive and inaccessible. However, there has been a significant effort of making available simpler and more accurate systems for gait analysis and classification. This dissertation reviews recent gait analysis and classification systems, presents a new database with videos of 21 subjects, simulating 4 different pathologies as well as normal gait, and also presents a web application that allows the user to remotely access an automatic classification system and thus obtain the expected classification and heatmaps for the given input. The classification system is based on the use of gait representation images such as the Gait Energy Image (GEI) and the Skeleton Gait Energy Image (SEI), which are used as input to a VGG-19 Convolutional Neural Network (CNN) that is used to perform classification. This classification system is a vision-based system. To sum up, the developed web application aims to show the usefulness of the classification system, making it possible for anyone to access it.
|
|---|