Desarrollo de un algoritmo de clasificación automática de gestos de la mano.

Abstract

Of superficial myoelectric signals, the classifier recognizes the gestures double tap, fingers spread, fist, wave in, wave out and relax. For which artificial intelligence techniques based on a CNN convolutional neural network are applied. To carry out the project, the EMG-120 database of the Escuela Politécnica Nacional is used, which contains records of myoelectric signals obtained through the superficial device myo armband located on the forearm, from which images of the gestures are generated to be able to store them. a dataframe structure to then preprocess, train the neural network model using the ResNet structure as a pretrained network, performing tests varying the number of layers of the network, validation was performed with 20% of the database, the system was has been evaluated using classification model performance metrics, obtaining good results with both preprocessed and non-preprocessed data. In the implementation of the classifier, use is made of free computational tools such as the Google Colab environment, Pandas for data manipulation and analysis, Fastai for training, and model validation.