The hard part
Why safety & security data is hard to label.
Where generic annotation vendors usually lose accuracy — and where a senior team makes the difference.
- Lab-to-site gap: public datasets miss the lighting, angles and people of your deployment.
- Small objects at distance — hard hats, vests, goggles — under dust and changing light.
- Rare events (intrusions, fire, falls) that are underrepresented in raw footage.
- Privacy requirements for footage of employees and the public.
Use cases
What we annotate for safety & security.
PPE compliance
Detection of hard hats, vests, goggles and gloves, with alerts for missing equipment.
Intrusion & perimeter
Virtual boundaries and object tracking for construction sites and restricted zones.
Fire & smoke detection
Early-stage flame and smoke detection in warehouses, industrial sites and open areas.
Crowd monitoring
Density and flow analysis for venues, stadiums and public spaces.
Mobile surveillance towers
Detection and behaviour analysis for portable, solar-powered camera towers.
Access control
Person and vehicle detection for gates, car parks and premium areas.
Annotation types
- Bounding boxes for people and PPE
- Tracking IDs for intrusion and behaviour
- Segmentation for smoke and fire
- Face and plate anonymization
Case study · Safety1st
A facemask detection model trained on public data failed in a glass-walled showroom. We built a dataset of ~15,000 photos from the actual devices; the retrained model reached about 90% precision in production.
Read the full storyFAQ
Safety & Security questions, answered.
Why not just train on public PPE datasets?
Public datasets are collected in controlled conditions and miss your site's lighting, angles and people. In our Safety1st project, an environment-specific dataset is what took the model to about 90% precision in production.
Can you annotate footage from construction sites?
Yes. We have prepared datasets for construction-site mobile surveillance towers, covering vehicles, equipment, people and PPE in dusty, changing light.
How do you protect people's privacy in surveillance footage?
All data is processed in the EU under GDPR, and we offer face and licence-plate anonymization as a separate service.
Further reading
From the BoBox blog.
Safety 1st – Facemask Detection Case Study
Safety1st's facemask detection worked in the lab and failed at a car dealership. BoBox rebuilt the dataset from ~15,000 on-site photos — and the model reached ~90% precision in production.
Mobile Surveillance System Powered by Computer Vision
How BoBox partnered with Safety1st to implement computer vision on mobile surveillance towers for construction site security — from dataset preparation to 90%+ precision.
Does Computer Vision Support Detection and Firefighting?
Computer vision is a powerful tool for many business cases, but also actively supports the work of public services — including fire detection, prevention, and operational firefighting support.
Have safety & security data that needs senior eyes?
Tell us about your dataset, model and timeline. A senior team member will reply.
