Triple Spectral Fusion Framework for Sensor-Based HAR
A new research paper proposes a triple spectral fusion framework for sensor-based human activity recognition (HAR). The framework addresses challenges in fusing heterogeneous data from Inertial Measurement Units (IMUs) and establishing long-term context correlations. It includes adaptive complementary filtering for noise suppression, graph Fourier domain filtering for dynamic heterogeneous graph nodes, and adaptive wavelet frequency selection to reduce context redundancy. The paper is published on arXiv with ID 2605.02743.
Key facts
- Paper proposes triple spectral fusion framework for HAR
- Uses IMU data including posture, motion, and context
- Adaptive complementary filtering for noise suppression
- Graph Fourier domain filtering for heterogeneous node fusion
- Adaptive wavelet frequency selection to reduce redundancy
- Published on arXiv with ID 2605.02743
- Addresses challenges in temporal information fusion
- Focuses on daily activity identification
Entities
Institutions
- arXiv