Authors : M.Robinson Jimenez, S.Oscar Aviles and Diana M. Ovalle
Abstract: This study presents two training schemes of three deep convolutional neural network architectures applied to object recognition, based on the depth information supplied for a 3D camera. For this case, the depth information allows to make the set of training images of each network, its architecture and its characteristics, generating a dynamic recognition application by variation of the image capture point. The best scheme is selected to add a weighting layer with saturationn for obtain a final architecture that recognize objects to different distances with a 91.69% success that mean a maximum error of 8.31%.
M.Robinson Jimenez, S.Oscar Aviles and Diana M. Ovalle, 2018. Convolutional Neural Network Training for Robotic Applications in 3D Environments. Research Journal of Applied Sciences, 13: 41-46.