Egocentric Activity Capture: First-Person Video Data Collection

This project collects real-world, first-person (egocentric) video of people performing everyday tasks, captured from head-mounted or chest-mounted cameras. The footage is used to train and evaluate models that need to understand human activity from the wearer's point of view: hand-object interaction, task sequencing, spatial awareness, and intent prediction.

Contributors record short, unscripted clips of themselves completing routine activities in natural settings: cooking, cleaning, assembly and repair, shopping, office work, tool use, and similar. Each recording follows a defined scenario brief covering environment, task, duration, lighting conditions, and camera placement, and is delivered with a metadata sheet (device, resolution, frame rate, location type, consent record). Where required, sessions include synchronized sensor streams such as IMU.

We are looking for resources who can support: participant recruitment, on-site or at-home recording, scenario scripting, and post-capture QC against the collection guidelines. Volume, target locations, and per-scenario requirements are defined per batch.