DAMIER

The illicit or malicious use of drones is a real threat that is currently only partially addressed by anti-intrusion systems. The DEEPLOMATICS project presented as part of ANR-18-ASTRID 0008-03 proposed an innovative, adaptive solution to the problem of intrusion detection on critical sites and infrastructures, both in open space and in urban areas. The proposed approach was based on innovative Deep-Learning techniques, enabling a scientific and technical leap forward in the identification and robust real-time tracking of drone trajectories. Within the framework of the DAMIER project, we propose to go beyond certain limitations of the technologies developed within the DEEPLOMATICS framework, by specifically addressing the problems of coordinated intrusions of several drones on sensitive sites. The themes of the DAMIER project are in line with the expectations set out in the Defence Innovation Orientation Document (DrOID) 2022, which aims to accelerate the development of anti-drone solutions. Artificial intelligence and autonomous systems, addressed in the DAMIER project, are also priority themes for DrOID 2023. The innovation-through-experimentation approach proposed in this study aims to improve operational performance both for the detection and localization of hostile UAVs, and for the autonomous deployment of an allied UAV for target-tracking tasks. The consortium partners are active members of the defense industry, such as Lerity, a specialist in vision systems, Acoem/Métravib-Defence, a supplier of acoustic detection systems in France and worldwide, and the fundamental research laboratories of LMSSC and ISL, which initiated the first project. The creation of a database of complex flight scenarios will complement the one produced by the initial project. This augmented database will then be used to improve the performance of the Beamlearning-ID neural network developed in the Deeplomatics project and integrated in collaboration with Métravib-Défense into an industrial prototype in the scope of the RAPID Deeplodocus. The aim here is to improve the detection range of a drone approaching the zone under surveillance, and to manage the localization of multiple sources using a single antenna. In addition, two imaging systems will be studied: the first will use Single Photon Avalanche Diode cameras which, in active mode, detect photons coming from an object in the scene and evaluate its distance. The second, a HEMISPACE system developed by Lerity, covers, day and night, a 360° field of view when fully equipped. It will be deployed here in a 120° restricted field-of-view variant, which will transmit real-time information on drone detection on the image, as well as information on acoustic detection of drones transmitted by the data fusion computer. A payload integrating a microphone antenna and a camera will be developed and fitted to an allied drone. It will embarq an autopilot controlled by the data fusion computer and the on-board computer. Detection and localization algorithms will be implemented to trigger emergency maneuvers in the event of a risk of collision or detection of a direct threat to the allied drone. Information transmitted by the sensors of all project partners will be integrated into the data fusion, along with a control capability for the allied drone and its payload. The performance of multi-source tracking will be evaluated by implementing algorithms such as JPDA (Joint Probabilistic Data Association) or Hungarian-net in the data fusion process.