Automated detection and classification of objects in the traffic flow on satellite images of the city

The paper discusses the developed methods of detecting and classifying objects in a traffic flow on ultra-high spatial resolution space survey data. Due to appearing the large amounts of free access satellite data, the development of machine learn-ing methods based on geospatial data, in particular satellite data, is becoming increasingly urgent. The paper justifies the choice of a source of data on traffic flows – ultra-high resolution satellite images. It also describes the main problems and tasks associated with the recognition and classification of objects in traffic flows. The purpose of scientific work is to develop and study a chain of algorithms that allows detecting and classifying objects in traffic flows with high accuracy. The research is based on a numerical assessment of the quality of the algorithms. The work uses the methods of pattern recognition, machine learning and digital image processing. The scientific novelty of the completed work is based on: a unique algorithm for extracting images of local sections of the road network, an algorithm for determining the direction of object's road movement, modernization of the selective search algorithm, which consists in filtering the extracted candidates. The work novelty is also confirmed by the fact that the used ultra-high resolution survey data have become accessible for private use relatively recently.