PPE compliance
Region-of-interest tags for PPE detection on industrial video: more than 3.4 million tags for a confidential EU client.
Problem
Workplace-safety AI must detect personal protective equipment and body regions on industrial video, so it can tell whether each person is wearing the right equipment in the right place.
Data and task
Region-of-interest tags across up to six body regions per person and more than nine PPE attributes, including colour classes. Each person is broken down by region, and each region carries the attributes the model needs.
Tools
Encord, used for tagging and review throughout the programme.
Challenges
Three things had to stay right across millions of tags:
- Pre-tagged boxes: building region tags on top of boxes that already existed.
- Attribute consistency: the same PPE item described the same way every time.
- Colour classes: applied reliably across different cameras and lighting.
QA
Two-tier review across the programme.
Delivery scale
More than 3.4 million region-of-interest tags for a confidential EU client.
Result and next step
A large PPE and body-region dataset for workplace-safety AI, delivered alongside the human pose keypoint annotation and workplace safety object detection projects for the same client.
Related industry: Industrial and workplace safety. Related service: Computer Vision.