Deep Learning for Ground Reaction Force Data Analysis: Application to Wide-Area Floor Sensing

Deep learning methods are proposed to process and fuse raw spatiotemporal ground reaction forces (GRF) to accurately categorize gait pattern. These methods are based on convolutional neural network and long short-term memory networks architectures to learn spatiotemporal features, automatically end-to-end from raw GRF sensor signals. In a case study on Parkinson's disease (PD) data, spatiotemporal signals of gait for PD patient and healthy subjects are processed and classified, resulting an effective gait pattern classification with a precision performance of 96%. Deep learning considerably achieved better classification results, compared to the shallow learning methods with the handcrafted features. This implies that for the purpose of automatic decision-making, it is beneficial to utilize deep learning methods to analyse GRF. This insight is portable across a range of industrial tasks that involve complex spatiotemporal GRF signals classification. The proposed models are computationally efficient and able to achieve high classification precision from a large set of GRF signals.