Abstract
Indoor positioning and navigation inside an area with no GPS-dataavailability is a challenging problem. There are applications such as augmentedreality, autonomous driving, navigation of drones inside tunnels, in whichindoor positioning gets crucial. In this paper, a tandem architecture of deepnetwork-based systems, for the first time to our knowledge, is developed toaddress this problem. This structure is trained on the scene images beingobtained through scanning of the desired area segments using photogrammetry. ACNN structure based on EfficientNet is trained as a classifier of the scenes,followed by a MobileNet CNN structure which is trained to perform as aregressor. The proposed system achieves amazingly fine precisions for bothCartesian position and Quaternion information of the camera.