Autonomous Drone Navigator for 3D Recon...

We propose a drone navigation and obstacle avoid-ance solution to support 3D reconstruction of unknown forestenvironments, using visual-inertial information and depth mapsas input. The navigation algorithm is built on Rapidly-exploringRandom Trees and incorporates obstacle structures and missionconstraints to enable 3D reconstruction by generating a path.Cameras mounted on the drone provide depth images that areused to extract point cloud information of the surroundingsand the ground slope. Using the point cloud information, a 3Dpath is computed utilizing onboard computational resourcestaking hardware limitations into account. The proposed solu-tion is independent of the GPS information for localizationand is capable of planning a smooth trajectory in real-time.Furthermore, the navigation algorithm also considers missionrequirements to avoid sharp turns, maintenance of acceptableflight speed, and distance to trees for accurate sensor datacollection. The solution is tested through simulations and isvalidated via experimentation.