Approach to robotic mobile platform path planning upon analysis of aerial imaging data

Abstract

The research is focused on solving one of the main problems of robotic system navigation—building an energy-efficient trajectory. A method of trajectory planning is proposed, according to which images obtained using an unmanned vehicle camera are stitched into an orthomosaic image, on which the Mask R-CNN neural network detects all static obstacles. In addition to creating a map of the area, the resulting images are also used to create a 3D model of the area. The developed method enables the robotic to find all static obstacles, as well as to identify all terrain features. Based on the obtained data on the state of the terrain and the presence of static obstacles, the system finds the most energy-efficient path.