The specific goal of the proposed project is to explore the applicability of selected advanced image processing technologies for the use in continuous urban expansion monitoring driven by high temporal density of images. Up to now, change detection approaches in EO are limited to typically bi-temporal analysis for two points in time. On the other hand, temporal trajectory approach assumes continuous seasonal development. The seasonal development approach is highly relevant for vegetated land covers, while urban expansion monitoring needs different algorithms.
Algorithms to process high-density temporal images are completely lacking especially for urban monitoring. Continuous image acquisition opens potential of reduction of the false change detection as well as potential of high automation in operational use. It is a challenge to separate variation between consequent images that represents changes of interest, which are typically fragments of the total area. The proposed approaches keep high potential of solving the currently existing constraints that are objective reasons for low automation of the change detection procedure, besides temporal limits.
The selected advanced technologies of image processing proved to be highly successful in other fields of application (such as bioinformatics, medicine, real-time tracking, document classification, gaming, etc.) The approaches to be analysed in the project are:
- Adaptive mixture models for modelling variations on image level;
- Markov Random Fields for modelling spatial context;
- Dynamic Bayesian Networks for representing priors for possible changes in the urban land-cover.