LETI Scientists Trained Drones to Navigate Autonomously Using Computer Vision
The development will reduce the risk of losing drones in the event of a failure of basic tracking systems.
Today, drones perform a wide range of tasks in a variety of areas of life, from delivery and crop irrigation to reconnaissance and high-rise diagnostics.
One of the problems associated with the use of drones (especially a group of drones) is ensuring their safety. This is especially true for complex multifunctional drones, the damage or loss of which can cause significant economic damage to the owner.
"We've developed software that allows the drone to use computer vision to navigate through the space and recognize the kind of terrain underneath it."
The development is based on a neural network trained on a wide range of information from an open database of video footage of different landscapes.
Thanks to the developed software, a drone can detect different types of terrain (water, vegetation, asphalt, and sand). Based on the analysis of the collected information, the drone can accurately determine its location. The results of the research have been published in the journal Symmetry.
"The fundamental importance of our work lies in the development of such an important field as computer vision for applications in various fields. Practically, the software can be used in hybrid mode with other drone tracking systems (such as GPS sensors) to reduce the potential risks of losing relatively expensive drones used in the agribusiness, for example."
Research team included Portuguese scientists. The project was carried out by specialists from the Laboratory for the Development of Systems for the Group Application of Land, Water, and Air Drones of LETI.