Saltar a contenido

Capture Protocol for Hand Grasp Data Collection

Date: 2026-05-07 Status: Initial version Refs: #34 (HAND-002)

Purpose

This protocol defines the procedures for capturing hand grasp images and video for the LABIS study. It covers equipment, setup, naming conventions, consent, and the distinction between synthetic and real data.

Equipment / Equipo

Minimum Required

  • Camera: Webcam (720p minimum) or smartphone camera with tripod mount
  • Tripod: Fixed position, adjustable height (50-80 cm from hand)
  • Background: Plain white posterboard or fabric backdrop (minimum 60x60 cm)
  • Lighting: Two diffuse light sources (desk lamps with diffusers) at 45-degree angles
  • Objects: As listed in object-grasp-camera-matrix.md per grasp type

Optional

  • Color calibration card for white balance consistency
  • Ruler or reference object in frame for scale calibration
  • Second camera for simultaneous multi-angle capture

Setup Procedure

  1. Mount camera on tripod at the distance specified in the matrix (distance_cm column).
  2. Position background behind the hand capture area.
  3. Set lighting to minimize shadows on the hand (diffuse, even illumination).
  4. Place a reference object in the first frame for scale (remove for subsequent captures).
  5. Verify camera focus and exposure on the hand at rest before starting.

MANDATORY for all real captures involving human subjects.

Before any real data collection: 1. Participant reads and signs the informed consent form (template in docs/templates/consent-form.md when available). 2. Consent form must specify: purpose of study, what body parts are captured, how data is stored, who has access, right to withdraw. 3. No faces should appear in captures. If face is accidentally captured, the frame must be cropped or discarded before storage. 4. Consent records are stored separately from image data, linked only by anonymous participant ID. 5. No real capture may proceed without documented consent.

Synthetic captures do not require consent as they involve no human subjects.

File Naming Convention / Nomenclatura

Pattern

{source}_{grasp_id}_{object}_{hand}_{angle}_{participant_id}_{sequence}.{ext}

Fields

Field Values Example
source synth (synthetic/AI-generated) or real (human capture) real
grasp_id From grasp-taxonomy.md power_grasp
object Object name, snake_case bottle
hand left or right right
angle dorsal, lateral, or palmar dorsal
participant_id Anonymous ID (P001, P002, ...) or synth P001
sequence Zero-padded frame/photo number 001
ext jpg, png, or mp4 for video jpg

Examples

real_power_grasp_bottle_right_dorsal_P001_001.jpg
real_power_grasp_bottle_right_dorsal_P001_002.jpg
synth_precision_pinch_coin_right_dorsal_synth_001.png

Synthetic vs Real Data / Sintetico vs Real

Synthetic (synth)

  • Generated by AI image tools or sourced from public domain datasets.
  • Purpose: pipeline validation only (smoke tests). NOT for scientific conclusions.
  • Must be tagged with source=synth in filename and manifest.
  • Stored in data/synthetic/ directory.
  • No consent required.
  • Limitations must be documented in any report using synthetic data.

Real (real)

  • Captured from human subjects following this protocol.
  • Purpose: scientific validation, model comparison, and clinical relevance assessment.
  • Must have documented consent.
  • Stored in data/private/ (not committed to git; tracked via manifest only).
  • Ground truth labels applied by trained annotator.

NEVER mix synthetic and real data in the same analysis without explicit separation and disclosure.

Capture Session Checklist

Before each session: - [ ] Equipment set up per Setup Procedure above - [ ] Background and lighting verified - [ ] Consent form signed (real captures only) - [ ] Camera settings noted (resolution, FPS, white balance) - [ ] Object set prepared per matrix row being captured

During capture: - [ ] One grasp-object-angle combination at a time - [ ] Hold grasp stable for minimum frames specified in matrix - [ ] Capture at least 2 repetitions per combination - [ ] Note any issues (occlusion worse than expected, object slipped, etc.)

After session: - [ ] Files renamed per naming convention - [ ] Manifest updated with new entries - [ ] Raw files backed up before any processing - [ ] Quality check: spot-review 10% of captures for focus, framing, and grasp correctness