Automatic Target Acquisition & Aim-Point Selection for Small Arms

The Army is seeking next-generation automatic target detection, recognition, identification, masking, tracking, and aim-point selection tools for small caliber weapons systems borne by dismounted soldiers to improve warfighter effectiveness. Accordingly, OKSI has proposed implementing state-of-the-art machine learning tools (e.g., deep learning) on ultra-low SWaP modern processors (e.g., Nvidia TX2) to deliver the above capabilities with SoA performance. During Phase-I, we implemented deep learning and other machine learning methods to provide real-time ATR, masking, target tracking, and masking for aim-point selection on military relevant targets at ranges out to 3 km using MWIR / LWIR / visible spectrum imagery sets (e.g., over 350,000 NVESD SENSIAC dataset frames). Our results showed that the deep learning methods proposed can achieve performance on-par with the well-known Johnson criteria for recognition and identification tasks. During Phase II, we will significantly enhance the initial algorithms we developed and will implement them for real-time aim-point selection and correction on an AIMLOCK small caliber weapon system in an operationally relevant scenario. Our work supports several DoD interests including (i) small arms aimpoint selection for guns and CROWS, (ii) improved ATR for seekers, rockets, and other munitions, and (iii) improved targeting and lethality for next generation combat vehicles.