Splatter SAR Image: 3D Reconstruction Using a Single SAR Image

In this paper, we introduce a new method to ex-ploring the conversion of single-view synthetic aperture radar (SAR) images to 3D object models using the Splatter SAR Image technique. The technique is based on Gaussion technology demonstrating an excellent ability to efficiently convert single- view SAR images into 3D models. We applied Gaussian Splatting to the 3D reconstruction process of single-view SAR images by training a neural network. The neural network operates in a feed-forward fashion during the testing phase. The network design is simple and uses 2D operators to map the input image to a 3D Gaussian distribution per pixel, thus generating a Gaussian ensemble that is rendered similarly to the image. To provide 3D information for model training, we make full use of existing millimeter wave radar data. This study not only successfully demonstrates the fast reconstruction capability of single-view SAR images, but also achieves better results in evaluation metrics. Ultimately, we demonstrate the significant cost and speed advantages of the method over other reconstruction techniques.