TITLE:
A Review of Pedestrian Recognition Based on Millimeter-Wave Radar and Video Fusion
AUTHORS:
Hongtao Li, Qimeng Lu, Huijia Gao, Beibei Xu, Zijia Chen, Zhengjie Wang
KEYWORDS:
Information Feature Extraction, Feature Fusion and Alignment, Cross-Modal Recognition and Matching
JOURNAL NAME:
Journal of Computer and Communications,
Vol.14 No.4,
April
22,
2026
ABSTRACT: Person identification serves as a core supporting technology in public security, intelligent home applications, and person tracking in different environments. Achieving stable identity matching across environments and locations has become a critical research requirement. Single-modal identification technologies suffer from inherent limitations. The cross-modal fusion method based on millimeter-wave radar and video realizes the complementary advantages of the two modalities. It can effectively support person identification across scenarios, becoming a research hotspot in complex open scenarios. This paper systematically reviews the research progress and core technical systems in this field. Firstly, it sorts out the information extraction methods of millimeter-wave radar features and video visual features, summarizing the extraction ideas of key features such as range-Doppler, micro-Doppler, point cloud, and skeleton. Secondly, it summarizes three fusion strategies (data-level, feature-level, decision-level) and two core alignment methods (spatio-temporal alignment, feature distribution alignment). Then, it introduces typical applications, including intelligent security monitoring, human-computer interaction authentication, and multi-target tracking. It analyzes current challenges such as complex environmental interference, cross-location spatio-temporal alignment, feature distribution shift, and scarcity of large-scale data. Finally, it discusses future research directions, including scene-invariant feature learning, cross-domain alignment optimization, unified feature space construction, and few-shot generalization, providing a comprehensive reference for person identification research.