Datasets:

xentim
/

fall-detection-sample

license: other
license_name: xentim-sample-evaluation
license_link: https://xentim.com/datasets/fall-detection/licence.txt
pretty_name: Fall Detection Dataset Sample
task_categories:
  - video-classification
tags:
  - fall-detection
  - pose-estimation
  - keypoints
  - skeleton
  - action-recognition
  - synthetic
  - cctv
  - computer-vision
size_categories:
  - n<1K
extra_gated_heading: Get the free Fall Detection Dataset sample
extra_gated_prompt: >-
  This sample is for internal evaluation only. See LICENSE.txt. Your details go
  to Xentim (TATO STUDIO sp. z o.o.), so we can reply about your evaluation.
  Privacy policy https://xentim.com/en/privacy-policy
extra_gated_button_content: Get the sample
extra_gated_fields:
  Company or institution: text
  Work email: text
  Role: text
  What will you test it on?: text
  I agree to use this sample for internal evaluation only: checkbox

Fall Detection Dataset v1.2: free sample

Free evaluation sample of a synthetic dataset for skeleton-based fall detection and person-down alarms.

Synthetic data: no real people and no personal data.

What is in this sample

336 sequences from 28 simulated events
data/frames.parquet: one row per frame and camera
data/index.csv: one row per sequence
data/skeleton3d/: 3D skeletons
preview_videos/: short skeleton previews
00_OVERVIEW_*.mp4: overview of the sample
DATASHEET.md: full specification and known limits

This is skeleton data, with no RGB video.

Load it

import pandas as pd

index = pd.read_csv("data/index.csv")
frames = pd.read_parquet("data/frames.parquet")
print(index.head())

Full dataset

12,108 sequences, 2.22 million annotated frames, 20.6 hours at 30 fps, 12 virtual cameras per event, 17 COCO keypoints, bounding box and fall phase per person per frame.

Commercial licence, from 18,000 EUR: Fall Detection Dataset

Licence

Free for internal evaluation only, see LICENSE.txt. Using it in a product needs a paid licence.

Questions or custom data: contact@xentim.com