Sledovanie tokov ľudí a dopravy na základe satelitných pozorovaní

Abstract

The dissertation focuses on the development and implementation of advanced methods for the detection of objects of interest such as cars and people using very high-resolution satellite imagery. With a current spatial resolution of up to 30 cm, these images allow for identification of smaller objects, which is essential for analysing a range of socio-economic phenomena, such as traffic, public gathering monitoring, or military conflicts. The main objective of this research is to define procedures for the detection and evaluation of individuals and vehicles using very high-resolution satellite imagery of urban environments. The analysis was conducted using data from WorldView 3 and 4 satellites capturing the Prague and Ostrava regions. Masks were generated using data from OpenStreetMap and digital technical maps of the cities. Object-Based Image Analysis (OBIA) and the Faster R-CNN machine learning model utilizing data augmentation techniques were applied in vehicle detection tasks. The detection accuracy (expressed by an F1 score) reaches 70 %. The results are negatively affected by shadows, overlapping trees, and horizontal road traffic signs. The detection of individuals was conducted using comparable methods. Within the OBIA framework, a semi-thresholding approach was employed, subsequently applying a local maximum to emphasize potential targets within the image. The accuracy of detection attained was up to 87 %. Conversely, the Faster R-CNN machine learning model, which incorporates data augmentation techniques, achieved an F1 score of 55 %, which is primarily attributed to the limited amount of training data available. The research demonstrates that the combination of traditional and modern image processing approaches can significantly improve the ability to detect objects in satellite imagery. These findings have the potential to be applied in numerous fields, such as analysis of social and economic trends, traffic management, security, and infrastructure monitoring.