Abstract: A optimal path planning device for drones, comprises: a first graph generation module configured to generate a first graph by setting a plurality of nodes in a predetermined area based on obstacle information and industrial structure information for the predetermined area, and by setting edges that connect each of the nodes; a second graph generation module configured to generate a second graph related to selected inspection object edges among a plurality of edges, a third graph generation module configured to generate a third graph with an Eulerian path based on the generated second graph, and a final path determination module configured to determine an optimal path of the drone based on the total cost of the generated third graph.

Abstract: An inspection system for a wind power generator includes a drone that transmits image information obtained by capturing images of a wind power generator and the surroundings of the wind power generator, and sensor detection information for detecting the wind power generator and the surroundings of the wind power generator; an inspection server that receives, from the drone, the image information and the sensor detection information as a transmission; and a mobile device that receives the image information and the sensor detection information, and controls operation of the drone by transmitting at least one instruction to the drone. On the basis of at least one piece of information from among the image information and the sensor detection information, the inspection server may identify a location of the wind power generator and inspect a current state of the wind power generator.

Abstract: An image stitching apparatus for inspecting a wind turbine, comprises a preprocessing module configured to generate a preprocessed image pair by removing background areas around blades of a wind turbine from photographed images of the wind turbine captured by a drone, an initial match point determination module configured to determine first initial match points in the preprocessed image pair by using a pre-trained deep learning module, a valid match point determination module configured to determine non-utilized match points out of the determined first initial match points, and determine valid match points by excluding the determined non-utilized match points from the determined first initial match points, and an image stitching module configured to perform image stitching between a plurality of images.