Use of semantic description of reference scene for visual navigation solutions

This article is devoted to the description generation of reference scene applying to visual navigation problem. The authors investigate the semantic descriptions that are more resistant to changing conditions of observation, thus being an alternative to existing approaches that use raster, feature, or object descriptions. There are no publications, considering similar approach of utilization for visual navigation problem solution. In the article the original description structure is proposed. The experiments involving descriptions of various scenes under various conditions have been conducted to assess the workability of the proposed approach. It is suggested that semantic descriptions are more resistant to partial occlusion of scene objects or changes in their visual features than other description types and can be used to solve visual navigation problems.