Considering camera distortion panoramic images forming method for unmanned aerial vehicle multispectral data

Abstract

The work is devoted to the study and modification of existing methods for merging a number of images obtained using a multispectral camera installed on an unmanned aerial vehicle into a single panoramic image for the purpose of its further thematic processing. A generalised method based on the existing detectors and descriptors of special areas of images was proposed for the automated solution of this issue, as well as the developed method for filtering matches of special areas. An analysis was carried out to select the best detectors and descriptors of special areas for the tasks of merging images of forest areas. It has been determined that the combination of ORB and FREAK methods show better results in detecting and describing specific points to perform the above tasks, than BRISK, SURF and ORB methods. Particular attention is paid to the importance of determining and correcting camera distortion used in data acquisition, the method used and the results of distortion correction are described. The effect of camera distortion to the displacement between singular points standard deviation in the case of data alignment is estimated. It is shown that the proposed automatic obtaining panoramic multispectral images method makes it possible to connect images with an average accuracy of up to 5 pixels when solving connection of multispectral images set issues.