Retail surveillance AI
Spatial and temporal annotation of in-store CCTV: more than 5,000 objects per batch on a private CVAT instance.
Problem
Retail intelligence needs both frame-level attributes and timestamped store events: what is in the scene, and what happens in it over time.
Data and task
Spatial boxes and behaviours on frames, covering people, baskets, phones and actions, plus temporal CCTV events with start and end times.
Tools
A private CVAT instance, rather than a shared workspace.
Challenges
Combining spatial and temporal annotation within rolling batches, so the boxes on each frame and the events on the timeline stay consistent with each other as new footage arrives.
QA
Two-tier review on every batch.
Delivery scale
More than 5,000 objects per batch for a confidential client.
Result and next step
Combined spatial and event data delivered in rolling batches.
Related industry: Retail. Related service: Computer Vision.