Video anonymization service
Video anonymization
for AI training data
We blur only faces, car plates and motorcycle plates, tracked through the entire video, so the scene stays usable for training. Every job is human-verified and delivered with an audit-ready completion report.
Why it matters
Privacy compliance
without losing training value
Real-world video captures people and vehicles who never agreed to appear in a dataset. Removing them is required, but blurring too much destroys the signal your models need.
Privacy rules apply to training data
GDPR, Korea's PIPA and similar laws treat faces and license plates as personal data. Datasets need documented de-identification before use or sharing.
Over-blurring breaks the data
Masking whole frames or wide areas hides road layout, body pose and interactions. Only identifiable regions should be removed.
Reviewers ask for proof
Customers and legal teams want to know how each file was processed, who checked it, and that the result was not altered.
Real result
Before and after,
side by side
Drag the handle to compare. Left is the original frame, right is the same frame from the final video processed by Origin V4. Roads and storefronts stay visible; only faces and plates are blurred.
OriginalOriginals are also masked for publication on this websiteOrigin V4 result
※ Originals are also masked for publication on this website
OriginalOriginals are also masked for publication on this websiteOrigin V4 result
※ Originals are also masked for publication on this website
※ Blue outlines mark blurred faces and green outlines mark blurred plates in the result images. Figures are taken from the completion reports.
Deliverables
What you receive
Each processed file comes with the evidence your compliance and data teams need.
Anonymized video
Faces and plates blurred and tracked across every frame.
Completion report
Processing summary, target counts and the full human review history.
Integrity hashes
SHA-256 of the original and result to show nothing else was altered.
Region log (JSON)
Machine-readable record of the regions that were blurred.
Engagement models
Three ways to work with us
Start with a free pilot on your own footage. Pricing and turnaround are confirmed after the pilot.
Processing service
- Send footage, receive anonymized video and reports
- Quoted per hour of video
- Data processing agreement included
With data collection
- Custom real-world data delivered already de-identified
- Review logs included in the dataset package
- One vendor from capture to delivery
On-premise deployment
- For teams that cannot send footage outside
- Runs offline on your machines
- Scoped case by case
Workflow
From pilot to production
Detection rules, review depth and acceptance criteria are agreed in the pilot before any volume work begins.
Free pilot
Up to 10 minutes of your footage, processed and reported.
AI detection & tracking
Faces and plates detected and tracked through each video.
Human verification
Low-confidence regions checked; full playback required.
Delivery
Videos, reports and logs delivered; source footage deleted.
Review tool
Built for human-in-the-loop review
Reviewers see blur applied live on the source video. The system flags only the regions the AI is unsure about, explains why, and will not mark a job complete until the full video has been played.
- Every edit and confirmation logged in time order
- Manual add, adjust and remove of blur regions
- Runs offline on a standard PC
Why Origin Data Lab
Built by a team that
produces AI training data
Origin Data Lab collects and processes real-world video for autonomous driving, robotics and Physical AI teams. The detection, tracking and human-review practices we use for our own datasets power this service.
Dense, messy footage
We work daily with crowded streets and markets full of people and vehicles in motion.
Dedicated review team
The same reviewers and procedures that verify our training data check every anonymization job.
Our own engines
Face, car plate and motorcycle plate engines developed and improved in-house.
| Targets | Faces · car license plates · motorcycle plates (additional regions can be added manually) |
|---|---|
| Method | AI detection → tracking through the video → human check of low-confidence regions → full playback before completion |
| Deliverables | Anonymized video · completion report (summary, target counts, review history, SHA-256) · blurred-region JSON log |
| Environment | Offline PCs · no external cloud upload |
| Input | MP4 and common video formats from vehicles, CCTV and handheld cameras (other formats on request) |
Our commitments
Your footage stays protected
These terms are written into our data processing agreement.
Offline processing
Footage is not uploaded to external cloud services for processing.
No model training
Never used to train AI models without your written consent.
Deletion after delivery
Source and results deleted once delivery is confirmed.
FAQ
Video anonymization questions
Compliance, data utility, deliverables, security and pricing.
01What does your video anonymization service do?
We detect and track faces, car license plates and motorcycle plates through the entire video and blur only those regions, so the scene, traffic and activity remain usable for AI training. A human reviewer confirms uncertain areas and must play the full video before a job is completed.
02Does it help with GDPR and other privacy regulations?
Our process is designed to support de-identification requirements under GDPR, Korea's Personal Information Protection Act and similar rules by removing identifiable faces and plates and documenting how each file was processed. Final compliance decisions remain with your legal or privacy team.
03Will blurring reduce the value of our training data?
We blur only faces and plates rather than whole frames or wide areas. Road layout, vehicles, pedestrians' body poses and scene context stay visible, which preserves most of the training value.
04What do we receive?
Anonymized video files, a completion report per file (processing summary, target counts, human review history, SHA-256 hashes of the original and result) and a JSON log of the blurred regions.
05Is the process fully automated?
No. AI performs the first pass, and a human reviewer checks low-confidence detections and plays the full video before completion. Difficult footage may need additional review, which we scope during the pilot.
06How is our footage protected?
Footage is processed on offline PCs without uploading to external cloud services, is never used to train AI models without your written consent, and is deleted after delivery is confirmed. Terms are set out in a data processing agreement.
07How is pricing calculated?
Pricing is quoted per hour of video, based on resolution, scene density and review depth. We confirm pricing and turnaround after a free pilot of up to 10 minutes of your footage.
08Can you anonymize data you collect for us?
Yes. Anonymization can be bundled with our custom real-world data collection so that datasets are delivered already de-identified, with review logs included.
09Can we run the software on-premise?
On-premise deployment is available on request for high-volume teams that cannot send footage outside. Contact us to discuss requirements.
Free pilot
Try it on up to
10 minutes of your footage
Tell us about your data. We will reply within one business day with secure transfer instructions and a data processing agreement. You receive anonymized files, a completion report and a region log before any commitment.
※ Please do not attach footage to this form. Video is transferred separately after the agreement is signed.
Startup programs
Selected for four global startup programs
Cloud Program