Video Annotation

Per-frame contours with instance identity held across the sequence, for objects whose shape matters – deformable items, angled vehicles, produce on a moving belt. Used where a bounding box captures too much background to be useful.

Frame-by-frame object annotation with a consistent instance ID maintained through occlusions, re-entries and scene changes. The highest-throughput video technique, used for detection, counting and trajectory analysis.

Cross-view identity matching across synchronised camera arrays, plus first-person footage from head-mounted and wrist-mounted capture. Instance IDs are reconciled across views before delivery.

Where the footage doesn’t exist yet, we capture it – scripted and unscripted, in field conditions, across our West Africa operation and partner networks. Includes egocentric and POV capture for robotics and wearable AI.

Skeletal and landmark annotation tracked across frames for human pose estimation, gait and biomechanical analysis, sports performance, and animal behaviour studies – with joint consistency maintained through partial occlusion.

Frame-level markers for discrete events – contact, phase transition, scene cut, anomaly, hand-off. Used in surgical phase recognition, sports analytics, manufacturing inspection and incident detection.