Génération de bases de données images IR sous contraintes avec variabilité thermique intrinsèque des cibles
This addresses the need for diverse and scalable training data in infrared imagery for ATR systems, but it is incremental as it builds on existing simulation techniques.
The paper tackles the problem of generating synthetic infrared image datasets for evaluating automatic target recognition (ATR) algorithms by simulating vehicle signatures on backgrounds with thermal variability, enabling the creation of large datasets.
In this communication, we propose a method which permits to simulate images of targets in infrared imagery by superimposition of vehicle signatures in background, eventually with occultants. We develop a principle which authorizes us to generate different thermal configurations of target signatures. This method enables us to easily generate huge datasets for ATR algorithms performance evaluation.