Patriot Park (Moscow region). August 28. INTERFAX-Specialists of the Military Space Academy named after Mozhaisk has developed a new algorithm for the operation of neural networks for drones, which allows identifying and tracking targets with high accuracy, the academy told reporters.
"A fundamentally new method of using artificial intelligence algorithms is proposed, which allows detecting an object, determining its coordinates and accompanying it during the entire shooting from a UAV. The algorithm is resistant to changes in illumination, as well as to changes in the shape of the object," the academy reported.
The algorithm consists in applying the principle of "super-resolution" by the neural network, when artificial intelligence creates three-dimensional models of any objects, for example, cars, in the smallest details with high accuracy.
At the same time, the representatives of the university stressed, the new algorithm provides the opportunity to work on low-performance boards.
"The system was tested in a real situation using a quadrocopter, which performed tracking of a moving car in an urban environment. Also, the algorithm of the neural network was tested on various video streams, " the academy said.
