Industrial Implementation and Performance Evaluation of LSD-SLAM and Map Filtering Algorithms for Obstacles Avoidance in a Cooperative Fleet of Unmanned Aerial Vehicles

Abstract

In this paper we present an industrial implementation and performance evaluation of the problem of obstacles detection by drones using autonomous navigation systems. The software module that has been developed as well as the tests conducted are part of a large industrial R&D Vitrociset project called SWARM: an AI-Enabled Command and Control (C&C) system, able to execute and review ISR missions for mini/micro cooperative fleets of heterogeneous UAVs. The presented software module, that is currently under test, has been developed to recognize obstacles and drive correctly the drones, using images acquired by low cost RGB video cameras, whose features of lightness and reduced size allow them to be installed on mini/micro UAVs. Moreover, this setup does not require special calibration and preconfiguration processes like the ones necessary for example using stereo video camera systems. The real-time recognition of obstacles in the surrounding environment has been obtained and evaluated through the implementation, performance evaluation and tests of the LSD-SLAM and map filtering algorithms; the core of the study has been realized starting from the integration of these algorithms with a simulated drone in a synthetic environment. The areas of interest have been identified through the filtering of a computer generated map: the module was then integrated into the SWARM project platform, allowing the control of a single drone's movement and making it ready for use in a cooperative fleet environment.