Authors : Javier O. Pinzon Arenas, Ruben D. Hernandez Beleno and Robinson Jimenez Moreno
Abstract: This study presents the training and validation of a deep convolutional neural network architecture used for a human-robot interaction. Two different datasets of images were employed with the aim of recognizing 2 kinds of hand gestures which are "closed" and "open" and control a robotic arm with these gestures. To choose the best training in the network, different behavioral parameters such as training accuracy and loss were evaluated to obtain the best training epoch and validation parameters such as validation accuracy and internal behavior of the network through the activations of the convolution layers. Once the trained network is chosen, camera tests and interaction with a robotic arm are performed, evaluating the interaction between the user and the actions of the robot through the network.
Javier O. Pinzon Arenas, Ruben D. Hernandez Beleno and Robinson Jimenez Moreno, 2017. Deep Convolutional Neural Network for Hand Gesture Recognition Used for Human-Robot Interaction. Journal of Engineering and Applied Sciences, 12: 9278-9285.