Video captureSetup 1Pilot-ready

Video Task Capture

A scalable, cost-efficient setup for collecting egocentric task video, workflow demonstrations, and structured episode metadata. Faster to deploy, easier to scale, ideal for rapid pilot programs across household, warehouse, and service environments.

Request a video capture pilotCompare: Glove capture →

Equipment

What the setup includes

  • Chest-mounted or head-mounted camera

    configurable per task

  • Optional hand-side view camera

    for hand visibility

  • Task logging via app or phone

    episode start / stop + metadata

  • Episode segmentation tooling

    automatic + manual override

  • Structured metadata per episode

    task ID, timestamps, labels, env

Best for

Ideal task types

  • Egocentric task video (POV)
  • Workflow and procedure capture
  • Repetitive physical task demonstrations
  • Success / fail episode labelling
  • Before-and-after task context
  • Multi-step household or warehouse tasks
  • Assembly-line or service workflows
  • Qualitative hand-presence detection

Industry applications

Where video capture is deployed

Video Task Capture works in environments where repetitive physical tasks can be observed from a first-person or near-POV perspective.

Cleaning & housekeeping

surface wipe, vacuum routes, object placement

Packing & fulfillment

box filling, label verification, parcel routing

Sorting & inspection

item classification, defect detection, bin routing

Service workflows

tray delivery, door handling, trolley push

Warehouse operations

shelf scanning, pick-and-place, inventory check

Household tasks

dish loading, laundry folding, table clearing

Data structure

Episode metadata schema

Each captured episode is delivered with a structured metadata record. Fields marked as buyer-specific are defined per program.

FieldTypeDescription
episode_idstringunique episode identifier
task_labelstringtask type from defined taxonomy
outcomeenumsuccess | fail | partial
duration_msintegerepisode duration in milliseconds
video_pathstringpath to primary video file
side_view_pathstring | nulloptional hand-side camera
environmentobjectlocation, surface type, objects
qc_statusenumaccepted | rejected | review
qc_notesstring | nullrejection reason if applicable
custom_fieldsobjectbuyer-specific metadata schema

Schema is extended with buyer-specific fields per program. Custom taxonomies, environment tags, and annotation layers are supported.

Tradeoffs

Advantages and limitations

Advantages

  • Lower equipment cost than glove-based setups
  • Faster contributor onboarding and training
  • Scalable to large contributor networks
  • Adaptable to many environment types
  • Ideal for rapid pilot programs
  • No wearable calibration required

Limitations

  • No fine-grained finger or joint data
  • Hand position inferred, not measured
  • Not suitable for dexterous manipulation modelling
  • Camera angle affects hand visibility

For fine-grained hand and finger data, see Glove Hand Capture.

Quality control

Default acceptance criteria

Each episode passes a QC review before inclusion in the accepted batch. Criteria are defined per program and can be extended with buyer-specific rules.

  • Camera stable and correctly oriented
  • Task fully captured start to finish
  • Correct outcome label applied
  • Required objects visible and identifiable
  • Metadata fields complete and valid
  • Episode duration within acceptable range
  • No obstructions blocking key action
  • Side view angle correct (if applicable)

Start a video capture pilot

Tell us the task type, environment, and episode volume. We will scope a pilot and propose a collection protocol.

Request a pilotSee Glove capture instead