Fall detection using low-resolution thermal array sensors (Melexis MLX90640 32×24, Omron D6T variants, Panasonic AMG8833). Currently in the dataset/research phase, evolving into a Python ML pipeline.
Dataset files (~3.5 GB) are not committed to this repo — they live on Google Drive:
https://drive.google.com/drive/folders/1R9WOARW9D-UJx-mKiefs8gLjRzWQRw2W
Shared with all Nami employees. Download the folder and place it as data/ at the repo root — the layout mirrors this repo's data/ exactly. See data/README.md for a per-dataset breakdown (sensors, licenses, file formats, fall counts).
To fetch the original public sources instead of pulling from Drive, run:
python scripts/download_datasets.pyscripts/download_datasets.py/scripts/download_*.sh— fetch each public datasetscripts/view_frames.py— animate a single thermal-array CSV as a heatmapscripts/dataset_viewer.py— interactive browser over all CSVs underdata/
| Dataset | Sensor | Resolution | License | Falls | Subjects |
|---|---|---|---|---|---|
| eHomeSeniors (2019) | MLX90640 + Omron D6T-8L | 32×24 + 1×8×4 | CC-BY 4.0 | 448 | 6 |
| Taramasco Multimodal (2025) | MLX90640 + radar + LIDAR + accel | 32×24 | CC-BY-NC 4.0 | ~100 | 10 |
| D6T-44L-06 GitHub repo | Omron D6T-44L | 4×4 | None | 238 | unknown |
| TADAR (2024) | MLX90640BAA | 32×24 | MIT | case study | 7 |
| Flier-01 (2023) | MLX90640 | 32×24 | None | 5,216 imgs | 2 |
Full details, file formats, and preprocessing notes: data/README.md.