Deep Learning Radar Point Cloud Input Improvement Method

Radar point cloud is used as the basic input data for deep learning, in order to improve the accuracy of radar data output, this paper will present the point cloud optimization algorithm for high-precision 4D radar, using the dynamic adjustment of spatial area echoes, and proposes a point cloud generation optimization method. At the same time, in the subsequent AOA angle resolution, echo space positioning can be further integrated with further matching optimization according to downstream needs. Finally, our detection data collected for the 4D radar in real road scenarios and results show that this 4D radar has significant performance improvements in terms of clutter suppression and density generation. It is known that the signal echo threshold is the product of spatial inversion after judgment. Therefore, in order to obtain a more accurate multi-channel the point cloud echo signal, the target detection of the echo needs to be optimized first.