Software and hardware bundle for controlling a group of unmanned vehicles based on robust computer vision algorithms
The widespread adoption of automated robotic systems requires their confirmed efficiency and safety. To test these criteria, experimental testbeds that simulate an external environment under the conditions similar to the operating environment are developed. Systems focused on performing tasks under real conditions should be robust against the dynamic external environment. In this work, we develop a combination of software and hardware components for controlling a group of unmanned vehicles (UVs) based on computer vision algorithms that would be robust to external environmental factors. As part of our investigation, UV prototypes and their testing environment are developed. The action decision-making of each UV is based on a computer vision system, which incorporates algorithms for positioning, movement correction, other vehicle and obstacle detection. To verify the developed system, we conduct an empirical study into how the UV decision-making is affected by such external environmental factors, as changes in the level and tone of lighting, as well as the appearance of the surface. The experimental results show that the developed and implemented UV computer vision system meets the established robustness and stability requirements regardless of the tested environmental conditions. The developed system minimizes deviations from the planned route, reduces the localization error to 10 % at most, provides the accuracy of 84 % of other UVs detection, and maintains UVs operation without failures for two hours. The developed hardware and software bundle and testbed can be used for security and smart city control algorithms verification, as well as for other tasks requiring UVs prototyping.
Authors: A. M. Belov, P. Yu. Belyaev, I. I. Viksnin, Yu. V. Kim, V. S. Radabolsky, T. A. Turushev, S. S. Chuprov
Direction: Informatics, Computer Technologies And Control
Keywords: unmanned vehicles, computer vision, neural networks, image preprocessing, robotic systems
View full article