Exploring Spatial-Temporal Representation via Star Graph for mmWave Radar-based Human Activity Recognition

2025-12-12

AI Summary

To address the challenges of sparse point clouds and variable-length sequences in millimeter-wave (mmWave) radar-based human activity recognition (HAR), this paper proposes an end-to-end framework integrating star-shaped graph modeling with a discrete dynamic graph neural network (DDGNN). We introduce, for the first time, a star-shaped graph structure wherein static human anatomical keypoints serve as global centroids and dynamic radar points as leaf nodes—explicitly encoding high-dimensional spatiotemporal relative geometry across frames. The lightweight DDGNN directly processes variable-length radar point cloud sequences without resorting to resampling or frame aggregation. Evaluated on a real-world mmWave radar HAR dataset, our method achieves 94.27% accuracy—comparable to vision-based skeleton methods (97.25%) and substantially outperforming three state-of-the-art radar-specific approaches. Ablation studies confirm the efficacy of both the star-shaped graph and DDGNN. Furthermore, the model has been successfully deployed on a Raspberry Pi 4 for edge inference.

Technology Category

• Planning, Routing, and Scheduling: Activity and Plan Recognition
• Machine Learning: Hardware-aware ML
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Human activity recognition (HAR) requires extracting accurate spatial-temporal features with human movements. A mmWave radar point cloud-based HAR system suffers from sparsity and variable-size problems due to the physical features of the mmWave signal. Existing works usually borrow the preprocessing algorithms for the vision-based systems with dense point clouds, which may not be optimal for mmWave radar systems. In this work, we proposed a graph representation with a discrete dynamic graph neural network (DDGNN) to explore the spatial-temporal representation of human movement-related features. Specifically, we designed a star graph to describe the high-dimensional relative relationship between a manually added static center point and the dynamic mmWave radar points in the same and consecutive frames. We then adopted DDGNN to learn the features residing in the star graph with variable sizes. Experimental results demonstrated that our approach outperformed other baseline methods using real-world HAR datasets. Our system achieved an overall classification accuracy of 94.27%, which gets the near-optimal performance with a vision-based skeleton data accuracy of 97.25%. We also conducted an inference test on Raspberry Pi~4 to demonstrate its effectiveness on resource-constraint platforms. sh{ We provided a comprehensive ablation study for variable DDGNN structures to validate our model design. Our system also outperformed three recent radar-specific methods without requiring resampling or frame aggregators.

Problem

• Addresses sparse and variable-size point clouds in mmWave radar HAR
• Proposes a star graph with DDGNN for spatial-temporal feature extraction
• Enables effective human activity recognition on resource-constrained platforms

Innovation

• Star graph representation for spatial-temporal feature extraction
• Dynamic graph neural network handling variable-size point clouds
• Resource-efficient inference on constrained platforms like Raspberry Pi