README.md

---
pretty_name: Human Motion Skeleton & Activity Dataset
license: other
task_categories:
  - time-series
tags:
  - computer-vision
  - human-pose-estimation
  - human-motion
  - activity-recognition
  - skeleton
size_categories:
  - n<1K
---
# Human Motion Skeleton & Activity Dataset — Public Evaluation Sample

Release: `human_motion_sample_2026_08_17_v2`

A lightweight public evaluation sample of anonymized human motion data extracted from CCTV/RTSP video and converted into structured skeleton pose and motion metadata.

This public release is intentionally limited to help researchers, ML engineers, and AI teams evaluate the usefulness of the dataset before requesting access to a larger commercial dataset.

## What's included

- 500 skeleton pose records
- 461 motion observations
- 17 motion segments
- Anonymous track IDs
- Time-aware motion sequences
- Automatically inferred activity labels
- Technical quality and provenance metadata

## Potential use cases

This sample may be useful for:

- Human pose estimation research
- Motion understanding
- Activity recognition
- Temporal movement analysis
- Skeleton-based machine learning
- Computer vision experimentation
- Privacy-preserving motion analytics
- AI model prototyping

## Source

- Source type: CCTV / RTSP
- Camera ID: `CAM-01`
- Source session: `20260816T102302Z`

The public dataset contains structured motion information only.

## Privacy

No raw CCTV imagery is included.

This sample does **not** contain:

- Raw CCTV frames
- Face crops
- Face images
- Face embeddings
- Private event clips
- Employee names
- Personal identity fields

The dataset is designed around skeleton-only / structured motion representations.

## Labels

Current motion labels are automatically inferred / pseudo-labeled.

They are **not human-verified ground truth**.

Current supported motion categories may include:

- standing
- walking
- running
- sitting
- bending
- falling
- unknown

See `docs/LABEL_DEFINITIONS.md` for details.

## Dataset files

### Data

- `data/skeleton_pose.jsonl`
- `data/motion_observations.jsonl`
- `data/motion_segments.jsonl`

### Metadata

- `metadata/sample_manifest.json`
- `metadata/source_manifest.json`
- `metadata/source_provenance.json`
- `metadata/source_quality_report.json`

### Documentation

- `docs/DATA_DICTIONARY.md`
- `docs/LABEL_DEFINITIONS.md`
- `docs/RIGHTS_AND_PROVENANCE.md`

## Sample limitation

This release contains a maximum of 500 skeleton records and represents only a small evaluation subset.

It should **not** be treated as the complete Human Motion Dataset.

## Larger dataset availability

Larger datasets with substantially more motion observations and sessions may be made available separately for evaluation, research, or commercial licensing, subject to applicable rights and redistribution requirements.

If your team is working on:

- Human motion models
- Pose / activity recognition
- Robotics
- Computer vision
- Behavior analytics
- Skeleton-based AI

and would like to evaluate a larger sample or discuss dataset requirements, please contact the dataset publisher through the platform profile.

## Release strategy

Public samples are expected to be updated regularly as additional validated motion data becomes available.

Each public release remains intentionally limited, while larger validated datasets are retained separately.

## Important notice

Technical provenance and quality validation do not by themselves establish legal redistribution rights.

Any larger commercial release is subject to appropriate rights verification and licensing terms.