| Title |
A Study on the Real-Time Moving Target Calculation Method Based on AI |
| Authors |
이민형(Min-Hyung Lee) ; 김도현(Do-Hyun Kim) ; 안희영(Hee-Young Ahn) ; 이영기(Young-Ki Lee) ; 한유근(Eu-Geun Han) ; 임소진(So-Jin Lim) ; 김경호(Kyung-Ho Kim) |
| DOI |
https://doi.org/10.5370/KIEE.2024.73.11.2011 |
| Keywords |
AI-based target estimation; CNN Model; DaSiameseRPN Algorithm; EO/IR camera imagery; Hardware Accelerator; Image processing; Leading point calculation; Lucas-Kanade Algorithm; Linear regression; Real-time target tracking; Motion Vector; Object tracking; Predictive tracking; Moving target analysis |
| Abstract |
Although various plans for future warfare are being discussed, no research has been published on a targeting device that predicts the aiming point of a target using the DaSiameseRPN algorithm, nor has any study demonstrated its real-time operation on a hardware accelerator. To process DaSiameseRPN in real-time on a hardware accelerator, an AI model that compensates for the accuracy loss caused by model compression was applied. This paper presents a novel approach to real-time target aiming by processing Image Tracer, DaSiameseRPN, and the Lucas-Kanade Algorithm on a hardware accelerator. The primary goal of this research is to accurately determine the aiming point of a moving target by leveraging advanced AI techniques in a real-time embedded system. To achieve real-time performance, we implemented fine-tuning, transfer learning, and model compression, enabling the AI algorithms to operate efficiently on the hardware platform.
The effectiveness of our approach was validated through comprehensive analyses, including motion vector extraction, linear regression analysis, and aim point error rate evaluation. The results demonstrate a substantial improvement in accuracy, with the AI system consistently predicting target positions with minimal error. Specifically, the integration of brightness clipping and linear regression led to a notable reduction in aiming errors, making the system more reliable in dynamic environments. Moreover, the system's ability to process complex AI algorithms in real-time on a hardware accelerator opens up new possibilities for deploying similar technologies in various real-world applications. The findings of this study confirm that the proposed method not only meets real-time requirements but also enhances the precision of target aiming, which is critical for applications such as defense systems, autonomous vehicles, and advanced surveillance systems. In conclusion, this research contributes to the field of AI-driven target tracking by providing a robust, real-time solution that can be implemented on hardware accelerators, thereby advancing the capabilities of current aiming technologies. |