01 — Data Type
Egocentric Video Data
Human-perspective video data for AI systems learning to understand activities, interactions and real-world environments.
What it is
Recorded from where the work happens.
Egocentric video is recorded from the human point of view — a camera worn at head or chest level, capturing a task the way the person performing it actually sees it.
That perspective matters because it preserves what a fixed third-person camera loses: where attention goes, how the hands approach an object, the order steps are taken in, and what the environment looks like from inside the task rather than across the room.
For systems learning to act in physical space, this is closer to the input the model will eventually receive than conventional footage is.
Why it matters for AI
Three things that decide whether it trains well.
- 01
The viewpoint matches the deployment
A robot or wearable sees the world from roughly where the camera sat. Third-person footage requires the model to bridge a gap that first-person footage never opens.
- 02
Hands and objects stay in frame together
Grasp, approach and release are visible as one continuous relationship, rather than reconstructed from a viewpoint that shows the person but not what they are touching.
- 03
Sequence is preserved, not implied
Real tasks have order, hesitation and correction. Captured continuously, that structure survives into the dataset instead of being flattened into a set of clips.
Applications
What this data trains.
Each of these draws on the category differently, which is why the specification is written per project rather than per service.
- Embodied AI
- Robotics
- Human Activity Recognition
- Spatial Intelligence
- Computer Vision
- Multimodal AI
- Human-Object Interaction
- Autonomous Systems
Collection possibilities
Where the data comes from.
Most engagements are a mix: annotate what you already hold, and collect specifically where coverage is thin.
Task-led collection
Collection is planned against the task families you name, not assembled from whatever footage exists. The brief specifies the activity, the environment and the conditions each recording must cover.
Environment and participant spread
Where generalisation matters, the same task is captured across different settings and different people, so the model does not learn one kitchen or one pair of hands.
Consent and provenance
Contributors record with informed consent, and each episode carries the metadata describing who recorded it, where and under what brief.
Annotation possibilities
Scoped to the project, not sold as a bundle.
These are the operations available for this data type. Which of them a project uses depends on what the model has to learn.
- 01First-person task capture
- 02Activity and sub-action segmentation
- 03Hand and object interaction labelling
- 04Environment and context metadata
- 05Multi-participant collection
- 06Structured episode delivery
Workflow
The same five stages, every project.
Understand
We start from the model and the use case, not the file format — what the system needs to learn, and what would make an example useless.
Specify
Requirements become a written collection or annotation guideline with worked examples and an acceptance definition, agreed before work starts.
Execute
Trained annotators and collectors work to that guideline, with edge cases escalated rather than guessed at.
Validate
Human review against the same rubric, plus consistency checks across annotators before anything is marked complete.
Deliver
Structured output in the schema your pipeline expects, with the guideline and review notes attached.
Quality workflow
Six gates between raw input and delivery.
The process does not change per data type. A corpus assembled across more than one category has to hold together, so the acceptance standard is shared.
Collection
Capture or sourcing against a defined specification, so the input is already in scope.
Guidelines
A written definition of correct, with worked examples and explicit edge-case handling.
Annotation
Trained annotators working to that guideline, escalating ambiguity rather than guessing.
QA
Structural and consistency checks across the batch, not just within single items.
Human Validation
Review by a second person against the acceptance definition.
Final Dataset
Structured delivery with the guideline and review notes attached.
Dataset preparation
What arrives at the end.
The output is agreed before work starts, so delivery is a handover rather than a translation exercise.
- Episodes delivered as continuous clips with their segmentation, rather than pre-cut fragments that lose context
- Task, environment, participant and session metadata attached per episode
- Annotation guideline supplied alongside the data, so your team can see what each label meant
- Structure agreed before delivery, in the schema your pipeline already reads
FAQ
Egocentric Video Data questions.
Video recorded from a first-person perspective, typically using a head or chest-mounted camera, showing a task from the viewpoint of the person carrying it out.
Have a data challenge?
Let’s build the dataset behind it.
Tell us what you’re building, what data you need and the scale of your project.