Computer vision for unmanned aerial vehicles (and GPS-free navigation) 2016
We will study a technology that enables to precisely navigate an UAV using only visual information. First, we compare current images from the camera to the set of reference images of the same landscape using phase correlation. Then we introduce altitude estimation with double and single camera (“cyclopus” approach). Next, we will get acquainted with mathematical aircraft modeling and usage of Kalman-like filters and recurrent networks for its control. Finally we will review some more complex task of UAV intelligence, as well as similar problems in on-ground navigation (cars).
Computer vision, unmanned aerial vehicles, phase correlation, altitude estimation, Kalman-like filtering, recurrent neural networks.
Python, Keras, Tensor flow, C++, Open CV, Matlab/Octave
Basics of computer vision and dynamic systems
Dr. Dimitri Nowicki
Computational neuroscientist–expert in AI and Neural Networks
Affiliation: Institute of Cybernetics of NASU, Ukraine/ Univ of Massachusetts CS dept.
Computational neuroscientist–expert in AI and Neural Networks – PhD (Université de Toulouse, 2004), MSc in applied mathematics, Moscow Institute of Physics and Technology (2000). Worked in France (Toulouse, Grenoble), USA (Univ. of Massachusetts, Harvard University). Associate Research Professor at the Institute of Cybernetics of NASU
Fields of interests: computational neuroscience, bioinformatics, recurrent networks and associative memory, optimization etc