A High-Accuracy human posture prediction based on Quasi-Distributed flexible FBG arrays and random forest algorithm

Abstract

Human posture prediction has practical value in numerous fields such as motion analysis, rehabilitation medicine, and human–computer interaction. This study proposes a novel approach for human posture prediction, realized through a quasi-distributed sensor array based on flexibly packaged fiber Bragg gratings (FBGs). During implementation, silicone rubber was employed to encapsulate the FBG sensors. The FBG array was strategically deployed within an insole according to foot pressure distribution patterns. By monitoring changes in FBG wavelength to track plantar pressure, precise prediction of human posture is achieved. Furthermore, a random forest (RF) algorithm combined with k-cross-validation (KCV) was employed to train on six static postures. The test subjects comprised 20 participants of varying ages and physical conditions. Experimental results demonstrate that the RF algorithm efficiently and accurately predicts the six static postures of test subjects. For individual subjects, the prediction accuracy, precision, and recall rates all reached 98.79%, with an F1 score of 0.99. For the entire group of 20 test subjects, the prediction accuracy, precision, and recall rates all reached 99.84%, with an F1 score of 0.99. The sensing system presented in this work demonstrates strong potential across multiple application domains, including diabetic foot evaluation, motor rehabilitation, virtual reality, and behavioral recognition, thereby enabling novel solutions and facilitating further advances in these fields.