Research Article
MM-HAR: Human Activity Recognition with mmWave Radar Micro-Doppler Features
@INPROCEEDINGS{10.1007/978-3-032-33662-0_34, author={Xiaohang Zhang and Yajun Zhang and Kun Liang and Chunhong Yue}, title={MM-HAR: Human Activity Recognition with mmWave Radar Micro-Doppler Features}, proceedings={Communications and Networking. 20th EAI International Conference, ChinaCom 2025, Xi’an, China, November 8--9, 2025, Proceedings}, proceedings_a={CHINACOM}, year={2026}, month={10}, keywords={Human Activity Recognition mmWave Radar Micro-Doppler Features Convolutional Neural Networks.}, doi={10.1007/978-3-032-33662-0_34} }- Xiaohang Zhang
Yajun Zhang
Kun Liang
Chunhong Yue
Year: 2026
MM-HAR: Human Activity Recognition with mmWave Radar Micro-Doppler Features
CHINACOM
Springer
DOI: 10.1007/978-3-032-33662-0_34
Abstract
Human Activity Recognition (HAR) has wide applications in scenarios such as smart healthcare and human-computer interaction. However, traditional vision-based methods often rely on lighting conditions and pose the risk of user privacy leakage. In contrast, millimeter-wave (mmWave) radar has gradually attracted widespread attention due to its contactless nature, insensitivity to lighting conditions, and strong privacy protection. This paper proposes a HAR system based on mmWave radar. First, the system combines Range-FFT and Doppler-FFT algorithms to extract micro-Doppler features from the echo signals. Subsequently, this paper proposes a lightweight model by integrating channel topology and Convolutional Neural Networks. In this model, efficient and lightweight local feature extraction is achieved through the proposed multi-channel branch convolution. Meanwhile, the model introduces dilated depthwise convolution to expand the receptive field, thereby maintaining lightweight characteristics and enhancing global perception capability. Experimental results demonstrate that the system achieves an average accuracy of 97.6% in HAR tasks, highlighting its excellent robustness and effectiveness, and providing strong support for practical applications.