Body Pose Analysis using CNN and Pressure Sensor Array Data

This paper addresses the machine learning techniques applied to analyze the human body positions "at rest" using a 3D body ‘impression’ data acquired by pressure sensors. The proposed experimental research used the "Bodies at Rest" dataset. The dataset was generated using pressure sensor arrays placed below a bedsheet. Unlike previous approaches, this paper proposes a model that classifies complex human poses using only pressure data. Eight poses were classified using convolutional neural network classifiers yielding an average precision, recall, and F1 score of 81.8%, 81.2%, 81.5% respectively. The ultimate goal is to apply the model developed in this study to assist with monitoring and analysis of motor function abnormality in hospitalized stroke patients using timeseries (sequential) pressure data. To do this three poses were classified using the model: supine, lateral, prone with recall scores of 99%, 92%, and 91% respectively.