COMPREHENSIVE METHODOLOGY FOR REAL-TIME PREPROCESSING OF UAV PHOTOREGISTRATION DATA

Abstract

The article analyzes the modern approaches used at the stage of pre-processing of unmanned aerial vehicle photo registration data by software algorithms. The development of the problem of efficient processing of large arrays of streaming data of photo registration systems of unmanned aerial vehicles in real time under conditions of limited computing resources is continued. The aim of the work is to build a holistic methodology for restoring and pre-segmenting the digital image matrix by software methods in accordance with the additive color scheme of the photo registration system without including methods for compensating for optical aberrations of the photo registration system and spatial filtering of input data. Methodology. An integrated approach to digital image processing is used, including the stages of pre-processing (PRIP), restoration and clustering of the image matrix with adaptation to the limitations of the computing resources of the hardware and software platform. Thresholding and morphological methods are used for processing, which provide preliminary segmentation and noise filtering, as well as classification and extraction of connected components in images. The scientific novelty is that an approach to the sequential application of erosion and dilatation procedures to remove noise, preserve the boundaries of important elements, and an algorithm for highlighting image boundaries due to the difference between the original image and the erosion result, depending on the color scheme of the image, has been developed. A three-stage approach for the color image segmentation algorithm is proposed, which includes detection of boundary gaps, thresholding, and region processing. It is proposed to limit the number of segments by introducing markers. Conclusions. The main attention was paid to the modeling and filtering of noise arising from the typical limitations of the photo registration system and the optical properties of the environment. The basic procedures of binary image preprocessing, such as dilatation, erosion, and edge detection, based on logical operators and correction primitives, were analyzed. The peculiarities of using thresholding methods of image preprocessing as a simple and effective approach for dividing an image into regions are presented. A scheme for implementing a color image processing algorithm with a global threshold is developed, which includes the stages of initial threshold estimation, correction, and re-checking until the specified accuracy is achieved. At the last stage, the methods of preliminary image segmentation using the watershed method are analyzed.