Describir: Indoor/Outdoor Semantic Segmentation Using Deep Learning for Visually Impaired Wheelchair Users