Launched at the end of 2023, the European FED STORE (Shared database for Optronics image Recognition and Evaluation) project aims to develop image recognition algorithms based on artificial intelligence, for better perception of land combat situations. The project includes the enrichment and sharing of an image database that will enable these algorithms to be trained and evaluated, the representativeness and variety of the images in the database being key factors in improving system performance.
The project provided an unprecedented opportunity to acquire operational images, with a campaign that took place from September 23 to 27, 2024 at the Fontevraud military camp, part of the Saumur Cavalry School. In order to have access to military vehicles whose signature could be shared by the various countries in the consortium, the FED STORE team approached the Musée des Blindés de Saumur, which made 6 armored military vehicles available for the week. Static and dynamic images in wooded and semi-urban environments were acquired, by chance, in a wide range of weather conditions (cloudy, heavily rainy, sunny and backlit) over the course of a week.
The campaign brought together 45 people from 9 of the project’s 21 partners (Thales, Safran, Onera, Marduk, Hensoldt, Rheinmetall, Konsberg, Flysight and Leonardo) for the deployment of some twenty ground-based optronic systems, as well as a dozen UAVs used for data acquisition but also as threats.
The volume of data acquired amounts to several hundred hours of recordings, annotated with observations to facilitate the search and indexing of sequences of interest. The next steps involve verifying the observations associated with the images, selecting the first data of interest to enable their annotation (in the AI sense of the term) and setting up foundation model annotation tools.
In order to vary the environments and conditions for creating the richest possible database, a second campaign is planned for this winter.