Videoloft’s PPE detection model
Our PPE Detection model is designed to improve construction site safety by automatically monitoring compliance with personal protective equipment – specifically the wearing of hard hats and high-visibility clothing.
The model detects people crossing a user-defined line in a video feed, typically an entrance to a safety zone.
Still images are captured and sent to our PPE detection model.
The model checks for high-visibility jackets and hard hats.
Results are displayed in the PPE dashboard, flagging compliance or non-compliance.
How its trained
- Thousands of cameras from Videoloft customers have been opted in to Videoloft’s analytics training programme, so the model is trained with real customer CCTV footage from construction sites globally – ensuring accuracy in real world angles, lighting, and environments.
- We also supplement this with AI-generated imagery to add edge cases, and expand the dataset.
Releases
Each release is tested extensively and benchmarked. Below you’ll find version history, accuracy scores, and known limitations.
V22 Initial release Aug 25 91% accuracy
Release Date August 2025
Summary First production release of PPE Detection Model.
Accuracy results
| Metric | Accuracy |
|---|---|
|
Full compliance detection (hard hat + high-vis) |
89% |
|
Missing hard hat detection |
93% |
|
Missing high-vis detection |
90% |
|
Full non-compliance detection (Missing hard hat + high-vis) |
92% |
|
Overall detection
|
91%
|
We test each version of our model against a set of real customer video and AI generated images. Let’s explain what these numbers mean…
- Of all the times model V22 detected full hard hat and high-vis compliance, it was correct 89% of the time.
- Of all the times model V22 detected a missing hard hat, it was correct 93% of the time.
- Of all the times model V22 detected missing high-vis, it was correct 90% of the time.
- Of all the times model V22 detected missing hard hat and high-vis, it was correct 92% of the time.
- Therefore the overall detection accuracy of model V22 is 91%.
Known limitations
There are some situations where the model has limitations. The example images below are AI-generated to protect customer data.
Camouflage-pattern hard hats not being detected as hard hats.
Neon sportswear being misclassified as high-vis jackets.
Workers carrying bulky objects (e.g. cables) obscuring visibility jackets.
hello@videoloft.net
Toll-free US: (855) 737-0237
UK: +44 203 9870133
+44 753 280 1187
Try Videoloft PPE detection for free
For more information or to get involved training our model, contact us directly or fill in your details using our quick form and a representative of our global sales team will be in touch.
Your details
We value your privacy and take data protection seriously. We will never share or sell your personal data to third parties. Your trust is important to us, and we’re committed to safeguarding your information.