Introduction

The Machine Perception Research Laboratory focuses on research and development of remote sensing data processing and visualization for more then 10 years. The key challenges addressed by the Laboratory are the following:
1. Automation: Current systems require deep human intervention and do not enable a fully automated processing of RS data. Nowadays the size of the databases and the continuously augmented content present a real barrier to most of the applications, since manual or semi-manual evaluation is expensive and time consuming.
2. Data heterogeneity: In complex Geographic Information Systems (GIS) space borne, airborne and terrestrial RS data is jointly utilized. On the other hand, there are several scanning procedures to produce RS data among others optical imagery (in visible or infrared spectrum), Synthetic Aperture Radar (SAR), TerraSAR and Inverse SAR (ISAR) imaging, Digital Evaluation Model (DEM) synthesis, or LIght Detection And Ranging (LIDAR) scanning. Real-world challenges need an efficient collaboration of many approaches coming from diverse domains. To enable such an integration, appropriate data description and handling models are required, which allow researchers to deal with sufficiently general and well-recognizable entities.
3. Various spatial scales of the data:  The quick quality improvements of the available data provide fundamental challenges for the processing modules. In the recent years, one of the main evolving features of the RS data is the spatial resolution. However, to interpret the content of the observed scene is often even more challenging than it was in the 2D case, since  one should reconstruct from large and irregular point clouds the appropriate region, object and sub-object structures.
4. Various temporal scales of the data: Change detection is an important part of many remote sensing applications. However, the definition of “relevant change” is highly task specific, leading to a large number of change detection methods with significantly different goals, assumptions and applied tools. Even for a given specific problem the data comparison may be notably challenging, considering that due to the large time lag between two consecutive image samples, one must expect seasonal changes, differences in the obtained data quality and resolution, 3-D geometric distortion effects, various viewpoints, different illumination, or results of irrelevant human activity.
5. Adaptation: The quick evolution of data quality and characteristics, as well as the availability of highly heterogeneous databases call for constant adaptation and innovation of methods and algorithms. To guarantee their smooth evolution, a sufficiently general and well-established data description and adaptively learning framework is needed.
6. Targeting the recently emerging needs of the applications: The high frequency refreshing of data in Geographic Information Systems (GIS) alerts in environmental surveillance systems, monitoring human offload or building activities need novel methods of analysis and evaluation.

Latest results