Monocular Vision Aided Drone Localization in Indoor Corridor

Abstract

Vision based navigation of unmanned aerial vehicles (UAV) has been an active field of research in the past decade. There are many challenges in making the vision system
understand the environment in which it is placed. Such an environment can be either indoors or outdoors depending on the task a hand. In This project we deal with localization of UAV in GPS-denied indoor environment. We consider A.R Parrot drone is our UAV model. Our aim is to localization of drone in corridor during flying time with the help of monocular camera which is attached with drone. In this project we designed navigation algorithm which will tell us the position of drone in corridor by seeing the images which are taken by drone camera. We create on dataset for localization algorithm by capturing the image of corridor for possible position of drone in corridor. Our localization algorithm implement with help of deep learning models use to perform regression task on the images which are taken by drone camera. We use help of transfer learning take recent deep learning models already trained with Imagenet Challenge data set .We modify deep learning models remove last layer and add own convolution and fully connected layers base on our desired output and train with our created dataset of corridor images.