Answers are grounded in this paper and cite their sources.
Human behavior understanding has broad applications across various domains such as human-computer interaction, autonomous navigation and surveillance. Human posture could offer rich behavioral information with small data burden, making accurate human posture estimation (HPE) the foundation of human behavior understanding. Currently, HPE methods are mainly based on visual modal data, and suffer from important challenges in visibility-limited conditions, such as low light, bad weather, and occlusions. Over the past years, microwave radar has been proposed as an effective method to solve the above problems, however, the substantial hardware cost has limited its application in many practical scenarios for accurate human posture estimation (HPE). In order to address these limitations, we based on intelligent metasurface with low energy consumption and low cost, complete microwave imaging pipeline for 3D HPE in this letter. In addition, we introduce a transformer-based diffusion model for microwave imaging, which can obtain stable and accurate results. Selected experimental results have been presented to demonstrate that our method is capable of achieving an HPE accuracy comparable to the visual route. Moreover, even in scenarios where the visual route fails completely due to occlusion, our method maintains great performance.
No takes yet. Share an insight, caveat, or question.
Bi et al. (2024) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: