Indoor Point Cloud Imaging With Millimeter-Wave Radar Based on Target Segmentation

To address the difficulty of accurately distinguishing static and moving targets with a single millimeter-wave radar in multitarget scenarios—which impacts self-velocity estimation accuracy—this article proposes a target segmentation-based millimeter-wave radar indoor point cloud imaging method (TSMIP). On our collected mmWave radar dataset, the proposed method achieves 83.89% segmentation accuracy, 39.42% higher than MobileNetV3 with 10 times fewer parameters. Compared to ResNet50, it is only 0.43% less accurate while reducing parameters by 100 times. Against the latest lightweight network, it cuts parameters by 46.44% with just a 0.23% drop in accuracy. The runtime of lightweight target segmentation network is reduced by 57% and 42% compared to ResNet50 and MobileNetV3, respectively. In addition, imaging results show that TSMIP maintains robust performance in environments with multiple moving targets. TSMIP is unaffected by the speed of moving pedestrians, ensuring stable, and accurate point cloud data. It avoids issues like scattering, which can degrade image quality. This technology is suitable for unmanned devices in smart industrial environments, where precise radar-based imaging is crucial.