High-Precision 3D Point Cloud Reconstruction and Annotation for Multi-Object Scenes

Using three-dimensional (3D) environmental simulation techniques to analyze electromagnetic (EM) environment distribution and predict EM radiation is an effective method for comprehensive and efficient EM management and spectrum optimization. Establishing a 3D model that closely aligns with the real environment is a prerequisite for accurate EM environment prediction and a critical factor influencing the accuracy of simulation results. However, existing modeling methods face challenges such as long modeling times, extensive manual operation, low model accuracy, and poor generalization. This paper proposes a high-precision method for 3D point cloud reconstruction and annotation for multi-object embedded scenes, utilizing a combination of millimeter-wave radar and visual sensors. This approach effectively distinguishes between objects that occupy similar spatial positions but are composed of different materials, making it suitable for multi-object scenarios. The generated 3D model, annotated with material EM parameters, achieves a labeling accuracy of 90.94%, thereby providing high-precision input models for EM simulation and prediction.