Multiple Object Scene Description for the Visually Impaired Using Pre-trained Convolutional Neural Networks.