Source time
Frames have timestamps.
Analyze source-time observations rather than confusing playback cadence, repeated decoded frames, or interpolation with independently captured evidence.
The Test Bench / Case study 03 / Temporal motion
Motion &
Interaction Bench.
Investigating tool movement, temporal sampling, camera artifacts, and possible contact events in a blacksmith video—without mistaking an image-plane estimate for physical ground truth.
RESEARCH CASE STUDY · No independently verified standalone Motion & Interaction Bench deployment. This page documents measurements; it is not a validated tracker.
Question / temporal evidence
More frames. Same failure?
Sampling faster can make a motion trace denser without making the tracker more correct. The relevant question is whether the source evidence supports the recorded positions and event interpretations.
Continuity
Tracks can be lost.
Hammer, hand, and workpiece positions can become ambiguous through occlusion, motion blur, exposure changes, and mistaken associations. Report gaps rather than invent continuity.
Camera effects
Appearance is not causality.
Color and apparent movement may change with exposure, viewpoint, or camera motion. A visible change is not automatically a strike or material deformation.
Independent review
Measure against evidence.
Compare generated tracks with independently reviewed visible positions and uncertainty labels, not merely with their own smooth-looking output.
Sampling density did not rescue the tracker.
The reported experimental comparison used a 0.12-second sampling interval (75 observations) and a 0.04-second interval (224 observations). The percentage of samples with usable tracking remained low or fell for the measured targets.
| Target | 75-frame pass (0.12 s) | 224-frame pass (0.04 s) |
|---|---|---|
| Hammer | 21.3% | 21.9% |
| Workpiece | 38.7% | 37.5% |
| Hand | 30.7% | 25.0% |
Increasing temporal samples did not meaningfully repair hammer tracking. The lesson is about measurement stability and failure attribution, not a certified statement about physical contact or a validated tracking algorithm.
A dense, smooth path can still be wrong. Track coverage, missed observations, source time, and reviewer-confirmed visibility matter more than presentation quality.
FRAME SAMPLINGTRACKING COVERAGECAMERA ARTIFACTSPHYSICAL CAUSALITY
Boundaries and related work
Not a finished motion tracker.
Image-plane trajectories and manual annotations can support a useful investigation. They do not establish force, true contact, 3D kinematics, or authentic physical history without additional evidence.
What remains unresolved
Long occlusions, camera-relative motion, stable object identity and accurate contact detection need independent ground truth. The blacksmith footage is a research case, not an accepted performance standard.
Related public instrument
Event Integrity Bench ↗ is a separate public alpha for annotation limits, manual event review, and mask/occlusion QA. It is not a standalone deployment of this Motion & Interaction experiment.