Accurate 3D reconstructions of objects are essential for several applications, from robotics, computer graphics and virtual reality to medicine, geology, agribusiness and architecture. Unmanned aerial vehicles (UAVs) have been increasingly used to capture aerial images as smaller and more accessible models become available on the market. These images can be used to generate high-quality 3D models of the overflying scene, but the quality of the resulting model depends significantly on the flight plan and the route executed, and still requires experienced pilots for complex environments. Thus, the research related to obtaining images for the reconstruction of large structures has been gaining importance, especially with regard to the perception capacity of these vehicles. In general, due to the lack of accurate (and current) information about the environment, and also for safety reasons, automated solutions with drones resort to flights in regular patterns at a safe air distance. However, the views captured are, in many cases, insufficient for high-quality 3D reconstruction. Thus, the transition from automated systems to autonomous systems is fundamental. In this scenario, this project proposes to explore a way to optimize the flight path in real time, based essentially on the vision acquired through an RGB camera attached to the robot, in order to improve the final reconstruction of the area of interest.
News published in Agência FAPESP Newsletter about the scholarship: