Multitemporal Data Analysis with Marked Point Processes

In this chapter, we introduce new approaches for object-level dynamic scene modeling based on multitemporal measurements, by extending the conventional Marked Point ProcessMarked Point Process (MPP) framework with modules focusing on the time dimension. First, a new probabilistic method is proposed for simultaneously extracting building footprints and performing change detectionChange detection in pairs of remotely sensed images captured with several years of time differences. The output of the method is a population of 2D building footprint segments, where status information is provided for each segment highlighting changes between the two time layers. In the second part, we propose a Multiframe Marked Point ProcessMultiframe Marked Point Process (F $$^m$$ m MPP) model of line segments and point groups for automatic target structure extraction and trackingTracking in Inverse Synthetic Aperture Radar (ISAR)Inverse Synthetic Aperture Radar (ISAR) image sequences. For the purpose of dealing with scatterer scintillations, and ensuring robustness despite the high level of speckle noise in the ISARInverse Synthetic Aperture Radar (ISAR) frames, we derive the output target sequence of the detector by an iterative optimization process, which takes into account in parallel the captured ISARInverse Synthetic Aperture Radar (ISAR) image data and different prior geometric interaction constraints between the fitted target samples of the consecutive frames. For both models, detailed quantitative evaluation is performed on real remotely sensed measurements.