Enhancing Static Activity Recognition Using Plantar Pressure Feature Engineering

This thesis is focused on how plantar pressure data can be used to improve the recognition of static activities, such as standing or sitting. The work is based on an existing activity recognition system that uses pressure-sensing insoles and machine learning models, and it focuses on the challenge of clearly distinguishing static postures from dynamic movements, especially when the available data are unbalanced. New features are extracted from the plantar pressure signals to better describe how pressure is distributed across the foot, including simple measures related to load distribution, center of pressure, sensor activation, and pressure differences between foot areas. To obtain reliable results and avoid biased evaluations, a leave-one-test-out validation approach is used. Different classification models are tested using the same preprocessing steps. The results show that the proposed features improve the recognition of static activities compared to the baseline approach, highlighting the value of plantar pressure information for this task.