Lower limbs motion intention detection by using pattern recognition

Electromyographic (EMG) signals processing allows to perform the detection of the intention of movement of the limbs of the human body in order to further use this decision to control wearable devices. For instance, robotic exoskeletons main objective consist of a human-robot interface capable of understanding the user's intention and reacting appropriately to provide the required assistance in an opportune way. In this paper, we study the performance of superficial EMG intended to design a intent pattern recognition based on Artificial Neural Networks (ANN) trained by the Levenberg-Marquardt method. Experiments consisting in 231 EMG records corresponding to 13 lower limbs muscles from 21 healthy subjects were considered. The EMG signals were randomly divided into the following sets: 70 % for training, 15 % for validation and 15 % for evaluation. The ANN-based pattern recognition was evaluated sample per sample with the movement intention annotations (target) and after the traininig operation end, the performance was evaluated in relation to the events (number of steps). The results show an accuracy of 90,96% sample per sample and 94,88% for an based on events evaluation. These findings motivates the use of this methodology for the classification of the motion intention detection in subjects with pathologies in the lower limbs.