HIATUS

Archival aerial images are a unique and a relatively unexplored mean for long-term environmental monitoring and change analysis. Indeed, the time series are long, dense temporally, acquired at very high spatial resolution. More remarkably, the survey conditions enable to generate height information. However, the time series heterogeneity and lack of reference information is limiting accurate dataset production and semantic information extraction. We believe these challenges can be overcome by exploiting the height information, the synergy between extraction of information and fine localisation, by using the entire long and dense time series and by taking into account data quality at each step. The HIATUS project is a collaboration between 5 laboratories with complementary methodological and thematic skills and a startup company. First, methods to improve and automatise fine georeferencing of archival image time series are proposed. To this end, we come up with the idea to identify dense and stable over time ground reference. Then, clustering and transfer learning techniques are expanded. For both of this work packages, the idea is to integrate spatio-temporal and thematic constraints to cope with data discrepancies. Finally, we propose to valorise the dataset and methods proposed. An on-line interface will be established to visualise the times series and the evolution of several thematic land cover use-cases. An on-line interactive thematic change detection tool will be developed. 3 PhD students, a 2-years postdoctoral are requested on the research work packages. A start-up company will contribute for valorisation (2 years). The total amount of budget is 535 keuro, divided in recruitment (465 keuro), equipment (8 keuro), trips (46 keuro), and additional expenses (16 keuro) (workshop organisation). With the management and structure costs, this amounts to a total grant of 578 keuro. We believe that the generic methods proposed will be useful in the image processing community to go from mono to multi-sensor image time series. The outcomes of the HIATUS project could have great impact in the human and social sciences because many challenges related to 2D/3D evolution of landscapes of the 20\textsuperscript{th} could be tackled. We will endeavour to distribute the outcomes of the HIATUS in both the scientific community and general public though research infrastructure and mainstream web interfaces.