Development of a Method for Accelerating the Processing of Streaming Video Received from UAVs

The article theoretically proves that when implementing stream processing at ground stations, the advantage should be given to methods of pipeline-parallel computing using cluster installations that can be equipped with graphics accelerators. The following results are obtained in the problems of detecting specified objects and zones of interest using the proposed pipeline-parallel implementation, which is solved by considering various configurations of cluster computing nodes (CCN): when processing full-color images of remote sensing (RS) on CCN with 5 dual-processor nodes, the maximum acceleration is 5.93 times compared to the score on one processor; predicted acceleration 10 times achievable by increasing the total amount to 8 nodes used CCN; when classifying 48 regions of interest on 20 multispectral images of RS on CCN with 16 nodes, an acceleration of 10.4 times was obtained in comparison with the counting at one node.