DeepMV: Multi-View Deep Learning for Device-Free Human Activity Recognition
summary
DeepMV is a multi-view deep model that fuses COTS Wi-Fi and ultrasound signals from 12 heterogeneous views (also evaluated with 9 Wi‑Fi views), trained on data from room1 and room2 and tested in room3. It builds a Feature Extractor with a Hierarchically-Weighted-Combination module and an adversarial domain discriminator to remove environment-specific features. Testbed uses 8 volunteers performing 8 daily activities plus 1 undefined activity; model consistently beats state-of-the-art baselines by a large margin.
AI-written summary · about
related talks
06mApproaching the Real-World: Supporting Activity Recognition Training with Virtual IMU Data
28mDeep Learning-Based Human Activity Recognition for Physical Skills...
06mUnsupervised Human Activity Representation Learning with Multi-task Deep Clustering
01h 23mMultimodal Deep Learning in the Real World
57mMachine Learning in the Age of IoT
06mReal-time Arm Gesture Recognition in Smart Home Scenarios via Millimeter Wave Sensing
27mDissecting ActivityRecognition with Multi-Sensor Simulation and Its Practical Deployment
01h 30mDeep Learning Techniques for Monocular Depth Estimation