Authors : M. Robinson Jimenez, S. Paola Nino, Oscar F. Aviles and Diana Ovalle
Abstract: This study presents the training of a convolutional neural network to identify different control signals made by hand, that allow to command a robotic mobile. Initially a database of 4000 images is established regarding the different control signals for the manipulation of the mobile, corresponding to 10 different users and after this the base structure of the convolutional neural network and the results of its training are determined. The robotic control algorithm was validated by means of navigation tests performed by 5 different users to those employed in the training stage where a percentage of accuracy was obtained to perform linear paths on average of 93.2% and for non-linear paths of 79%. Training algorithms for convolutional neural networks have not been evaluated in robotic navigation control tasks for transporting objects.
M. Robinson Jimenez, S. Paola Nino, Oscar F. Aviles and Diana Ovalle, 2018. Convolutional Neural Training for Robotic Control Through Hand Gestures. Journal of Engineering and Applied Sciences, 13: 8949-8954.