Target Radar Cross Section Prediction Method Based on Improved Generative 3D Model

To address the issue that traditional single-view generative 3D models cannot provide the absolute scale of a target, this paper proposes an improved generative 3D model method. When the target is within the common viewing area of an electro-optical (EO) camera and radar sensors and is moving rapidly, making it impractical for the EO camera to continuously capture multiple views of the target, this paper improves the traditional single-view generative 3D model algorithm. By utilizing the single target view captured by the EO camera combined with the target ranging information provided by the radar, it can generate a 3D model of the target with absolute scale. Compared to existing methods that combine radar and EO cameras to recover the absolute dimensions of monocular depth maps, the depth map absolute dimension recovery method can only yield non-closed meshes, providing dimensions solely for the side of the target facing the camera, and thus cannot be used for subsequent calculations. In contrast, this paper recovers the absolute scale of a complete 3D model, resulting in a fully enclosed mesh. Based on the acquired 3D model of the target with absolute scale, this paper proposes an engineering application that addresses the issue of insufficient prior information on target scattering characteristics during radar detection in complex environments, which leads to suboptimal radar tracking performance. By combining the obtained 3D model with radar cross-section (RCS) rapid calculation software, the radar can quickly acquire multi-angle RCS information of the target after detection. Compared to the statistical Swerling model, the multi-angle RCS information of the target provides precise and comprehensive radar characteristic information in all directions, offering greater practicality and accuracy for observing various types of targets. Based on this information, the radar can predict the variation of the target's RCS with changes in attitude, adjust the detection threshold in real time, and achieve stable tracking. Simulation comparison experiments demonstrate that using the target 3D model generated by this paper for RCS calculations yields reliable results with short computation times, making it applicable to practical engineering scenarios.