Use of Spectral Landscape Indices for Obstacle Detection in the Tasks of Mobile Robotic Platforms Navigation in Agricultural Areas

Abstract

Listen

Translate article

Purposeor research is to develop an algorithm for detecting obstacles on the orthophotomap based on the analysis of the spectral landscape indices in the tasks of mobile robotic equipment navigation in agricultural areas.Methods.The following landscape indices characterizing objects of various types on a map obtained by spectral aerial photography have been considered in the paper: normalized difference vegetation index (NDVI), normalized building difference index (NDBI), normalized difference water index (NDWI), and soil-adjusted vegetation index (SAVI). These indices provide an assessment of the four main classes of objects on the map: vegetation, buildings, water bodies, and soil cover. An algorithm that provides the segmentation of zones on the map which are impassable for ground robotic means using multispectral images and the considered indices was proposed.Results. Each image is presented in the form of a colour map based on the pixel-by-pixel calculation of the indicated indices. In this case, three indices, i.e. SAVI, NDWI, NDBI, are combined (superimposed on each other), and then the NDVI layer is subtracted from the resulting image to highlight the passable zones. Thus, a formula to obtain a mask of obstacles in the image was obtained. Hence, this algorithm allows generalizing the results of calculations for all selected indices and constructing a mask of obstacles in the image. For quantitative assessment the of the algorithm execution, the area of obstacles was calculated using the indices on a sample of manually marked images. The experiments conducted show that the developed algorithm provides, on average, detection of 85.47 % of the area of all impassable zones in the images in the above classes of land cover.Conclusion.An algorithm for the automated detection of obstacles on a map obtained from a spectral orthophotomap of the area for use in the tasks of mobile robotic equipment navigation in agricultural areas has been developed and tested. In the further research, to determine flat soil areas, it is planned to modify the developed solution using the improved modified soil-adjusted vegetation index (MSAVI).