3DCFAR-Net: A Coherent Accumulation Network for Target Information Mining on Millimeter-Wave Radar

Millimeter-wave (MMW) radar is a crucial component of advanced driver assistance systems (ADAS) and intelligent transportation systems, thanks to its affordability, continuous operational capability, and resilience in adverse weather conditions. Point cloud imaging is one of the most important techniques that enables a thorough perception of the surroundings and enhances target detection and recognition under practical scenarios. However, conventional radar point cloud imaging methods encounter performance limitations due to inadequate target reflectivity under complex electromagnetic environments. This study investigates coherent and non-coherent accumulation methods for point cloud imaging and introduces a novel neural network method, i.e., 3DCFAR Net, to overcome these limitations. The robustness and effectiveness of 3D CFAR Net are rigorously evaluated through both numerical simulations and real-world experiments under complex electromagnetic scenarios.