MmMask: A Human Mask Generation Method Based on MmWave Radar

Human contour segmentation serves as a fundamental task in computer vision, playing a critical role in image analysis. However, vision-based segmentation methods often raise privacy concerns and suffer from significant performance degradation in low-light environments. To address these issues, this paper proposes a novel method for generating human masks based on point cloud features extracted from mmWave radar. Specifically, the proposed approach first processes radar raw data using signal processing algorithms to extract the point cloud features of the human target. Then, leveraging the U-Net framework, a radar-based 2D human mask generation model is constructed. To enhance feature extraction, the U-Net blocks are designed using convolutional neural networks (CNNs) combined with Convolutional Block Attention Modules (CBAM). Finally, the method is validated using real-world measurements. Experimental results demonstrate that the proposed approach can effectively generate 2D masks from radar point cloud data, achieving an average accuracy of 94.56% on the test set.