Using Autonomous Drones Interactions towards Mobile Personal Spaces for Indoor Environments
Eric Marin Velazquez, Sudhanshu Kumar Semwal · Computer Science Research Notes · 2021
We propose an extension of a recent work using convo-lutional neural networks and drones, such as Learning tofly by using DroNet [8] that can possibly safely drive adrone autonomously. The combination of (i) the DroNetarchitecture and weights to apply to CNNs to avoid thecrashes; (ii) combining it with DLIB tracker, a corre-lation implemented tracker based on Danelljan et al.’spaper [3] work; (iii) the extraction of descriptors usingSpeeded Up Robust Features [1]; and (iv) Fast Libraryfor Approximate Nearest Neighbors [10] for the featurematching – leads a drone to track any object and avoidcrashes autonomously without any prior informationabout the object. The main goal is to create a partnershipbetween the drone(s) and the participant as the dronefollows the participant and avoids collisions. Our workextends existing methods to also included a way for adrone to follow a person even if the person is hiddenfor a few frames. Our algorithms also work in low/poorambient light satisfactorily. In future, our technique canbe used to provide novel indoor applications for drones.