Table 1.
Comparison of sensor-based HAR frameworks and methods.
| Helmy and Helmy [45] | Añazco et al. [46] | Köping et al. [47] | Mairittha et al. [48] | Gudur et al. [38] | Xia and Sugiura [25] | Kim et al. [49] | Ijaz et al. [50] | Google AR API [51] | Apple Motion [52] | Microsoft Kinect [53] | HARE (Ours) | ||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Data Collection | External Sensor Connectivity | ✗ | ✓ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ |
| Multimodal Recording | ✗ | ✗ | ✓ | ∼ | ✗ | ✗ | ✗ | ✓ | ✗ | ✗ | ✗ | ✓ | |
| Extendability * | ∼ | ✗ | ✓ | ✓ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | |
| Classification | Real-time | ✓ | ✓ | ✓ | ∼ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✓ |
| Multimodality | ✗ | ✗ | ✓ | ✗ | ✗ | ✗ | ✗ | ✓ | ✗ | ✗ | ✗ | ✓ | |
| Labeling | Refinement | ✗ | ✗ | ✗ | ✓ | ✗ | ✗ | ✗ | ✗ | ✓ | ✓ | ✗ | ✓ |
| Anonymization | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✗ | ✓ | ✓ | ✓ | ✓ | |
| Sensor Placement | Pose Estimation | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✓ | ✗ | ✓ |
| Optimal Position | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | |
| ML-based Feedback | On-Device Training | ✗ | ✗ | ✗ | ✗ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ |
| Human-in-the-Loop | ✗ | ✗ | ✗ | ✓ | ✓ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | |
| Extended Logging | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✗ | ✓ | ✗ | ✓ | |
∼: Partially supported. *: Adaptable to different domains through extending the landmark model. : Additional statistics per person and activity.