Discrete classification of upper limb motions using myoelectric interface

Electromyografic signals offer insights into understanding the intent and extent of motion of the musculoskeletal system. This information could be utilized in developing controllers for applications such as prostheses and orthosis, and in general assistive technology. This paper presents a myoelectric based interface to control five discrete upper limb motions involving the shoulder and elbow joint. Four subjects performed the experiment, which consisted of two separate phases: the training and testing phase. Extreme Learning Machine algorithm is used to classify the myoelectric signals to the control motions. The data collected during the training phase is used to train the parameters of the decoder, and the data from the testing phase is used to quantify the performance of the decoder. The muscle activations of each subject are used to manipulate a virtual human avatar. The graphical visualization serves to provide real-time feedback of the motions generated. The performance of the decoder for both offline and online classification are evaluated. Results indicate an overall classification accuracy for online control being 78.96±23.02%. The rate of transition from rest phase to the desired motion phase, on an average is 0.25 ± 0.10 seconds.