Recognition Delay and Recognition Rate of Knee Motor Intention Recognized by Electromyogram and Continuous Hidden Markov Model

Abstract

A motor rehabilitation robot applied patient’s intention can enhance the rehabilitation efficacy. Continuous hidden Markov models of knee flexion and extension are trained using autoregressive model coefficients of knee flexor and extensor electromyograms. The patient’s intention of knee movement are recognized by the trained continuous hidden Markov models and the user’s knee flexor and extensor electromyograms. The suggested method was applied to a knee joint rehabilitation robot for identifying the suggested classification method in real time. A nondisabled healthy subject wore the robot, and its knee joint was extended when the subject’s intention was recognized as ‘Extension.’ The robot’s knee joint was bended when the subject’s intention was recognized as ‘Flexion’. If the user’s intention wasn’t recognized as ‘Extension’ nor ‘Flexion’, the robot’s knee joint was remained stationary. The robot had followed properly the subject’s knee joint motor intention. As a result of hidden Markov model classification, the robot reflects the subject’s intensions with the recognition delay shorter than 200 msec and the recognition rate of 94.23 %. The results show the suggested method has good potential as a bio-signal classification method for a motor rehabilitation robot.