Analysis of Matrix Factorization Techniques for Extraction of Motion Motor Primitives

Through the study and analysis of motor primitives, it is possible not only to perform motor control to aid impaired people, but also to develop techniques for neuromuscular rehabilitation. Human motor control is executed through a basic set of signals that govern the motor behavior. This basic set of signals is called primitive motor movements. From the knowledge of an individual’s motor primitives, it is possible to develop control strategies, which are capable of assisting individuals with some sort of mobility impairment. In order to extract these motor primitives in a precise way to carry out the development of assisting devices, factorization techniques are fundamental and many methods exist to perform this task. The present work analyzes the results yielded by four of the most used matrix factoring techniques (PCA, ICA, NNMF, SOBI) using electromyography signals. The results suggest that the PCA is the technique that best managed to reconstruct the EMG signals after their factorization, with a virtually zero relative error. The SOBI technique also yielded satisfactory results, followed by NNMF and finally ICA, which presented a reconstructed signal quite different from the original.