ctechvent/arisole-strideiq-gait

Arisole StrideIQ Gait Dataset — v0.1 Seed Release

Paper: Toward an Open, Privacy-Preserving Benchmark for Markerless Gait Analysis
Authors: Varun Srivastava, Arisole
Version: 0.1 (Seed Release — aggregate statistics, no per-clip data)
License: CC BY 4.0
DOI: (pending Zenodo assignment)

---

Dataset Summary

The Arisole StrideIQ Gait Dataset is the first open, smartphone-native, markerless gait-analysis benchmark derived from real-world walking sessions captured entirely on consumer hardware — no wearables, no motion-capture suits, no laboratory infrastructure required.

v0.1 (this release) publishes aggregate platform statistics from 100+ consented walking sessions captured using the Arisole StrideIQ mobile application. It establishes the data schema, biomechanical pipeline, and baseline distributions ahead of a planned v0.2 clip-level benchmark.

Why This Matters

Existing gait datasets (e.g., CASIA-B, CMU MoCap, DIRO) require laboratory hardware costing tens of thousands of dollars. StrideIQ captures equivalent biomechanical signals from a standard smartphone camera, opening gait analysis to:

• Low-resource clinical settings globally
• Remote rehabilitation monitoring
• Robotics locomotion research (Open X-Embodiment compatible schema)
• Population-scale epidemiology

---

Dataset Structure

v0.1 Contents (Aggregate Statistics)

Field	Description
session_count	100+ walking sessions
landmark_schema	33 MediaPipe BlazePose landmarks (x, y, z, visibility)
biomechanical_features	Stride length, cadence, step symmetry, trunk sway, knee flexion angle
confidence_gating	Frames with landmark visibility < 0.6 excluded
MII_score	Movement Intelligence Index — composite gait health score (0–100)
platform_distributions	Aggregate histograms of MII and feature distributions

Coming in v0.2

• Clip-level landmark sequences (with explicit participant consent)
• Demographic metadata (age band, self-reported health condition category)
• Per-session biomechanical time-series
• Leaderboard for automated gait quality prediction

---

Biomechanical Pipeline

Smartphone Video
      │
      ▼
MediaPipe BlazePose (on-device inference)
      │  33 landmarks @ 30fps
      ▼
Confidence Gate (visibility threshold 0.6)
      │  Filters occluded/low-quality frames
      ▼
Biomechanical Feature Extraction
  ├─ Stride length estimation (hip–ankle geometry)
  ├─ Cadence (step frequency via heel-strike detection)
  ├─ Step symmetry (L/R stride time ratio)
  ├─ Trunk sway (shoulder midpoint lateral displacement)
  └─ Knee flexion angle (thigh–shin vectors)
      │
      ▼
Movement Intelligence Index (MII)
  Weighted composite score — normalized to 0–100
  Higher = more efficient, more symmetric gait

Key architectural insight (P2 companion paper): The full biomechanical pipeline runs entirely on-device. The landmark payload transmitted to the cloud is 130 KB/session vs. ~24.3 MB for the source video — a **200× compression** with zero raw biometric video leaving the device.

---

Usage

from datasets import load_dataset

# Load v0.1 aggregate statistics
ds = load_dataset("ctechvent/arisole-strideiq-gait")
print(ds)

Note: v0.1 contains aggregate statistics and the data schema. Clip-level data will be added in v0.2. Subscribe to / watch this repository to be notified when v0.2 is released.

---

Related Resources

Resource	Link
Companion Paper (P1)	Toward an Open, Privacy-Preserving Benchmark for Markerless Gait Analysis — Zenodo DOI (pending)
On-Device Architecture (P2)	The Server Doesn't Need to See You Move — Zenodo DOI (pending)
Global Health Context (P3)	Movement Poverty — Zenodo DOI (pending)
Arisole Website	arisole.com
arXiv Preprint	Pending endorsement (arXiv submission ID: 7864429)

---

Comparison to Existing Gait Datasets

Dataset	Capture Method	Cost	Clip-Level	Smartphone	Open License
CASIA-B	Multi-camera lab	$$$$	✅	❌	❌
CMU MoCap	Optical MoCap	$$$$	✅	❌	✅
TUG Dataset	Clinical sensors	$$$	✅	❌	Varies
StrideIQ (this)	Smartphone	$0	v0.2	✅	✅ CC-BY

---

Data Collection & Privacy

All sessions captured via the Arisole StrideIQ app under informed consent. v0.1 contains aggregate statistics only — no per-user or per-clip data is included. The raw video never leaves the participant's device (on-device inference architecture).

Institutional review: (IRB/ethics review in progress for v0.2 clip release)

---

Citation

If you use this dataset in your research, please cite:

@dataset{srivastava2026strideiq,
  title     = {Arisole StrideIQ Gait Dataset (v0.1 Seed Release)},
  author    = {Srivastava, Varun},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/ctechvent/arisole-strideiq-gait},
  version   = {0.1},
  license   = {CC-BY-4.0}
}

Companion paper citation:

@article{srivastava2026openmarkerless,
  title   = {Toward an Open, Privacy-Preserving Benchmark for Markerless Gait Analysis},
  author  = {Srivastava, Varun},
  journal = {arXiv preprint},
  year    = {2026},
  note    = {arXiv submission ID: 7864429, pending endorsement}
}

---

Maintainers

Varun Srivastava — Founder, Arisole
Arisole / Ctech Ventures
Contact: move@arisole.com | arisole.com

Feedback, collaboration requests, and robotics/rehabilitation research partnerships welcome.