Physical activity recognition via minimal in-shoes force sensor configuration
We propose a new minimal wearable system and a classifier for physical activity recognition. The configuration is solely based on two force sensors placed anteriorly and posteriorly under the feet. To find the optimal sensor configuration, we estimated the total force under the feet during daily activities. The estimation was based on a linear regression model built upon the forces estimated over selected areas from the dense mesh of high-resolution sensors of a commercially-available force sensing system. The best estimate of the total force, which also indicated the best sensor configuration, was fed to the activity recognition algorithm to provide the final output. The analysis indicated that the optimal locations which allowed estimating the total force with a minimal RMS error (40N) were the central part of rear foot and forefoot. Using this configuration and the activity classification algorithm, the classification accuracy for the basic activities such as sitting, standing and walking were 93.8%, 99.5% and 93.4%, respectively. These values demonstrate the high accuracy of the proposed system and are very encouraging for recognition of additional types of activities of daily-living in the next stage.