UAV Pose Estimation in Indoor Corridor Using Monocular Vision and Deep Learning

Abstract

Unmanned Aerial Vehicles (UAV’s, commonly referred as drone) in recent times have acquired large attention in research community. Due to their ability of agility, drones have got potential industrial or civil task.The control command for any UAVs in transit depends on the several factors, for e.g.position, velocity and orientation of the vehicle. These factors specification are not always available directly. Majority of the past research measured these factors using dedicated sensors and sophisticated hardwares. We proposed a model that uses only vision cue to automate the pose correction of drones in indoor flat corridor. The intrinsic features of the environment of the transit, earlier obtained by dedicated hardwares, in our case is obtained from the image of the environment using deep, feed-forward artificial neural networks called as convolutional neural network. Our approach is effective enough to enable any (robotic autonomous) device, to localize itself to face straight when it is at the centre of the navigating environment.