Myoelectric Algorithm for Knee Angle Estimation Using Proprioceptive Data and a Compatibility Test

Abstract

This article presents a method to estimate the knee angle based on data fusion for transfemoral leg prostheses control, using information from two electromyographic signals, two gyroscope sensors and one electrogoniometer channel. This information is processed in three stages: feature extraction using cepstral coefficients and the myoelectric signal entropy, pattern classification using a perceptron neural network and data fusion from a Kalman filter. A compatibility test is introduced based on Mahalanobis distance with the aim to detect possible artifacts to come from the estimated angle at the neural network output. The method was tested in healthy subjects, and the results were compared with another work that was based solely on myoelectric signals. The results showed that the use of additional information related to proprioception improves the precision of the knee joint angle estimation, and reduces artifacts.