Object-Grasp-Camera Matrix¶
Date: 2026-05-07 Status: Initial version Refs: #34 (HAND-002), depends on grasp-taxonomy.md (#33)
Purpose¶
This matrix defines every combination of grasp type, object, hand, and camera angle that the LABIS hand recognition study will capture. It serves as the blueprint for both synthetic smoke tests and real data collection sessions.
How to Read¶
Each row is one capture scenario. A complete data collection session should cover all rows. The success_criterion column defines what constitutes a valid capture for that scenario.
Matrix¶
grasp_id,object,hand,camera_angle,background,lighting,distance_cm,expected_occlusion,min_frames,ground_truth,success_criterion
power_grasp,bottle,right,dorsal,plain_white,diffuse,60,low,30,manual_label,21_keypoints_detected_confidence_gt_0.5
power_grasp,bottle,right,lateral,plain_white,diffuse,60,moderate,30,manual_label,finger_flexion_visible_in_keypoints
power_grasp,cylinder,left,dorsal,plain_white,diffuse,60,low,30,manual_label,21_keypoints_detected_confidence_gt_0.5
power_grasp,hammer_handle,right,dorsal,plain_white,diffuse,70,low,30,manual_label,21_keypoints_detected_confidence_gt_0.5
precision_pinch,coin,right,dorsal,plain_white,diffuse,50,low,30,manual_label,thumb_index_tips_close_lt_30px
precision_pinch,coin,right,lateral,plain_white,diffuse,50,moderate,30,manual_label,thumb_index_convergence_visible
precision_pinch,small_cube,left,dorsal,plain_white,diffuse,50,low,30,manual_label,thumb_index_tips_close_lt_30px
precision_pinch,pen_tip,right,dorsal,plain_white,diffuse,50,low,30,manual_label,thumb_index_tips_close_lt_30px
spherical_grasp,tennis_ball,right,dorsal,plain_white,diffuse,60,moderate,30,manual_label,5_fingertips_spread_around_object
spherical_grasp,tennis_ball,right,lateral,plain_white,diffuse,60,high,30,manual_label,thumb_abduction_visible
spherical_grasp,apple,left,dorsal,plain_white,diffuse,60,moderate,30,manual_label,5_fingertips_spread_around_object
lateral_pinch,key,right,dorsal,plain_white,diffuse,50,low,30,manual_label,thumb_tip_near_index_lateral
lateral_pinch,key,right,lateral,plain_white,diffuse,50,moderate,30,manual_label,thumb_press_against_index_side_visible
lateral_pinch,credit_card,left,dorsal,plain_white,diffuse,50,low,30,manual_label,thumb_tip_near_index_lateral
tripod_grasp,pen,right,dorsal,plain_white,diffuse,50,low,30,manual_label,3_tips_converge_thumb_index_middle
tripod_grasp,pen,right,lateral,plain_white,diffuse,50,moderate,30,manual_label,tripod_formation_visible
tripod_grasp,pencil,left,dorsal,plain_white,diffuse,50,low,30,manual_label,3_tips_converge_thumb_index_middle
tripod_grasp,stylus,right,dorsal,plain_white,diffuse,50,low,30,manual_label,3_tips_converge_thumb_index_middle
hook_grasp,bag_handle,right,lateral,plain_white,diffuse,70,low,30,manual_label,fingers_hooked_thumb_not_opposing
hook_grasp,bag_handle,right,dorsal,plain_white,diffuse,70,moderate,30,manual_label,pip_dip_flexion_visible_no_thumb
hook_grasp,bucket_handle,left,lateral,plain_white,diffuse,70,low,30,manual_label,fingers_hooked_thumb_not_opposing
Column Definitions¶
| Column | Description |
|---|---|
| grasp_id | Identifier matching grasp-taxonomy.md |
| object | Physical object used in the capture |
| hand | Which hand performs the grasp (left/right) |
| camera_angle | dorsal (back of hand), lateral (side), palmar (palm side) |
| background | Scene background (plain_white for controlled captures) |
| lighting | Lighting condition (diffuse = even, no harsh shadows) |
| distance_cm | Camera-to-hand distance in centimeters |
| expected_occlusion | Predicted finger/object occlusion level (low/moderate/high) |
| min_frames | Minimum video frames or photo count per scenario |
| ground_truth | Labeling method (manual_label = human expert annotation) |
| success_criterion | What the model output must show for this scenario to pass |
Coverage Summary¶
- 6 grasp types x multiple objects = 21 total scenarios
- 2 camera angles per grasp minimum (dorsal + lateral)
- Both hands represented (right dominant, left for variety)
- Controlled conditions: plain white background, diffuse lighting, fixed distance
- All distances 50-70 cm (typical webcam/tripod range)
Notes¶
- This matrix covers controlled conditions only. Field/ecological captures would need additional rows with varied background and lighting.
- The
success_criterionvalues are designed for automated pipeline validation, not clinical conclusions. - Real captures require consent per capture-protocol.md.