How egocentric video is captured: cameras, rigs and quality
Where the camera sits and how it is set up decides whether footage is usable for training. A buyer asks how a set was recorded before asking anything else, so a seller should be able to answer in a few lines.
Where the camera sits
| Rig | What to know |
|---|---|
| Head-mounted camera | Closest to the eyes, so the view matches what the wearer sees. It moves with every turn of the head. |
| Chest mount | Steadier than a head mount and keeps the hands in view for desk and bench work, but the view sits lower than the eyes. |
| Smart glasses | Look natural and sit at eye level, with limited battery and storage. Quality varies by model. |
| VR or mixed-reality headset | Cameras on the headset see the room, and many headsets track the hands, which gives hand pose without extra labeling. |
| Phone on a head or chest mount | Cheap and easy to roll out to many people. Quality depends on the phone and on how steady the mount is. |
| Screen recorder | For software work. It shows clicks and menus. Customer data on screen is masked first. |
What makes footage usable
- Hands stay in frame for most of the task
- A steady picture, without heavy shake or motion blur
- Good light, so objects and fingers are clear
- Enough resolution and frame rate to follow fast movements
- A wide enough field of view to see the work area
- Audio or narration when the buyer asks for it
- No dropped or corrupted files
Extra signals some rigs record
Some setups record more than video: depth, motion sensors that show how the head moves, where the wearer is looking, and tracked hand positions. These help a model but also add files and privacy questions, so a seller should list what each clip carries.
What to keep with each clip
Task name, a code for the wearer rather than a name, the type of place, the date, the device, the language of any narration and a link to the consent record. Without this a clip is a picture with no context. The data collection guide covers the consent side.
A quick check before you offer or request a set
Write down the rig, the resolution and frame rate, the hours, the number of different people, the tasks and settings, and whether there is audio or narration. If you are selling, that is your summary. If you are buying, it is the start of your request.
Read next
Questions about how egocentric video is captured: cameras, rigs and quality
Which camera is best for egocentric video?
There is no single best one. A head mount matches the eyes, a chest mount is steadier, and a headset gives hand tracking. What matters is that hands stay in view and the picture is steady and clear.
Can a phone be used?
Yes, on a stable head or chest mount. Quality varies, so a buyer will ask for a sample before accepting a phone-recorded set.
What resolution and frame rate do labs want?
It depends on the model and the task. Fast hand movements need a higher frame rate. A buyer states its needs in the request.
Does the set need audio?
Only if the buyer asks for it. Audio can capture other people's voices, so it adds consent and privacy work.
Do I need to keep the original files?
Yes. Keep the originals and a record of any processing, so a buyer can see what was changed.
Didn't find your question? Ask us directly.