What can you anonymize?

Faces, license plates, people's bodies and silhouettes, ID badges and cards, screens and monitors, street signs and addresses, and readable text on documents, in photos, video, scanned pages and screenshots.

Which anonymization methods are available?

Gaussian blur, pixelation, solid masks, synthetic face replacement, silhouettes or skeletons, and OCR-based text redaction. We recommend a method per use case, and you can mix methods by object class.

Can you keep data useful for AI training?

Yes. Synthetic face replacement and silhouettes preserve pose, expression and behavior, so models can still learn from the footage without learning who is in it.

Do you work with CCTV, dashcam and body-camera footage?

Yes. We handle varied resolutions, night footage, fisheye lenses and long recordings, and tune detection to each camera type.

Can you process live streams?

Our core service is batch processing of recorded footage with human review. For near-real-time needs we can design a pipeline with automated redaction and sampling-based review. We scope it case by case.

How accurate is detection?

Accuracy depends on footage quality, so we measure it on your sample and report recall for each object class. We run detectors at conservative thresholds and use human review to catch what automation misses.

How do you handle people who turn away or are partly hidden?

We combine detection with tracking across frames, so a person is covered even when the face is briefly hidden, small, blurred or at the edge of the frame.

Is blur safe, or can it be reversed?

Weak blur or large pixel blocks can sometimes be partly reconstructed. We choose blur strength and block size to resist this, and recommend solid masks or synthetic replacement for high-risk material.

Can anonymization be reversible?

Yes, if you need it for investigations. Identities are replaced by tokens and the originals are encrypted with keys you control. Each restore is logged. For datasets and public release, we recommend irreversible methods.

How do you deal with metadata?

We strip or rewrite EXIF data, GPS coordinates, device identifiers and embedded thumbnails, so identifying information does not remain hidden in the file.

How is our footage kept secure?

Encrypted transfer, isolated review environments with no downloads, vetted and NDA-bound staff, full audit logs and a short, agreed retention window followed by verified deletion.

Does this help us comply with GDPR and similar laws?

It can help you align with data-minimisation and privacy-by-design principles under laws such as GDPR, India's DPDP Act, CCPA and HIPAA. Whether data counts as anonymous is a legal question, so we work alongside your counsel and document our methods.

Who holds the original files?

You do. We process them in an isolated workspace for the agreed time, deliver results to your storage and then purge the originals, issuing a deletion record.

Can we choose where data is processed?

Yes. We can process in the region you require, or inside your own cloud environment, to meet residency requirements.

Do you provide proof of what was done?

Yes. Every batch ships with an audit report: what was detected, what method was applied, who reviewed it and what changed.

How is pricing structured?

Typically per hour of video or per thousand images, depending on content density, required accuracy and review level. After the pilot we provide a fixed-scope quote.

How long does a pilot take?

About one working week from kickoff to anonymized sample and accuracy report, once we have your sample footage.

How fast can you process large volumes?

Throughput depends on volume and review level. We plan capacity with you up front and add reviewers and compute to meet your delivery dates.

What affects cost the most?

Footage density, such as crowded scenes versus empty roads, the number of object classes, required recall, and whether you need full manual review or sampling.

Do you offer ongoing support?

Yes. Many clients run continuous programs with scheduled deliveries, shared dashboards and monthly quality reviews.

Image & Video Anonymization

See everything. Identify no one.

Aarvexa finds and permanently hides faces, license plates, screens and readable text in photos and video, so you can share, publish and train AI on visual data without exposing the people in it.

Faces & plates
Human-verified QA
Secure, audited

Get a free anonymized sample See how it works

RECcam-04_main-street.mp400:00:14:00

AnonymizedOriginal

0faces

0plates

0text & screens

Method

Blur · mask · synthetic

Metadata

EXIF & GPS stripped

Redaction methods

Six ways to make a person disappear

Different data needs different protection. Hover a card to see each method applied, or let the demo cycle.

Applied

Gaussian blur

A smooth, natural-looking blur. The most common choice for video and street imagery.

Best forStreet view, broadcast, CCTV exports

Applied

Pixelation

Mosaic blocks that clearly signal redaction. Block size is tuned to resist reconstruction.

Best forEvidence review, press, FOI releases

Applied

Solid mask

Opaque bars or shapes. The strongest visual guarantee, nothing underneath survives.

Best forLegal, medical and high-risk material

Applied

Synthetic replacement

Faces replaced with generated, non-existent faces that keep expression and pose for AI training.

Best forADAS, retail analytics, ML datasets

Applied

Silhouette & pose

People reduced to shapes or skeletons, keeping motion and behavior, dropping identity.

Best forBehavior analytics, safety studies

Applied

Text & document redaction

OCR finds names, numbers and addresses, then redacts them consistently across frames and pages.

Best forClaims, forms, dashcams, screens

What we detect

If it can identify someone, we find it

Toggle each class to see what the detectors look for. Every class gets its own model, threshold and review rule.

detections.preview8 objects found

Faces
Front, side, partial, small and occluded faces, including in crowds.
License plates
Plates from many countries, angles and lighting conditions.
Signs & addresses
Street names, house numbers and readable text in the scene.
ID badges & cards
Name tags, lanyards, access cards and printed documents.
Screens & monitors
Content on laptops, phones, kiosks and shop displays.
Also: tattoos and unique clothing, vehicle logos, documents on desks, whiteboards, voice tracks and EXIF or GPS metadata, on request.

The process

From raw footage to a clean cut

Six stages, every one logged. Hover a frame to pause the reel.

FRAME 01

Secure ingest

Files arrive over encrypted channels into an isolated workspace. Nothing is copied to personal devices.

FRAME 02

Detect

Models find faces, plates, text and screens in every frame, tuned on your own footage.

FRAME 03

Track

Each object is followed across frames, through occlusion, motion blur and re-entry.

FRAME 04

Anonymize

The agreed method is applied consistently: blur, mask, pixelate or synthetic replacement.

FRAME 05

Human QA

Trained reviewers check every flagged and low-confidence frame, and sample the rest.

FRAME 06

Deliver

Clean files and an audit report go to your storage. Originals are purged on schedule.

Accuracy you can tune

We choose missing nothing over looking pretty

Detection is a trade-off. Slide the confidence threshold and watch what is caught, what is missed and what gets blurred by mistake. Figures are illustrative.

Face coveredFace missedFalse alarmCorrectly ignored

100%Recall

88%Precision

Confidence threshold0.35

ours

Every face caught, little noise

All 44 faces are covered and only 6 false alarms remain, which human QA clears quickly.

Industries

Wherever a camera is pointed at people

Six sectors that rely on visual data and cannot afford to expose the people in it.

Automotive & ADAS

Make driving datasets shareable with partners and annotators without exposing pedestrians or plates.

Retail & smart stores

Analyse footfall and dwell time from in-store cameras while keeping shoppers anonymous.

Smart cities & CCTV

Release footage for planning, research and public requests with people and plates removed.

Healthcare & research

De-identify clinical photos and procedure video for teaching, publication and AI development.

Media & broadcasting

Protect bystanders and minors in news, documentary and user-generated content.

Insurance & legal

Share claims photos and evidence with all parties while redacting everyone else.

Dashcam dataFleet recordingsTraining sets

Reversible or permanent

Pick how final the anonymization should be

Some data must never be recoverable. Some must be, under strict control. Try all three.

Protected

AUDIT LOGNot applicable

Irreversible

Irreversible anonymization

Identity is destroyed permanently. No key exists, so the result is safe to share, publish or use for training.

Preferred for datasets and public release
Strongest privacy guarantee
Often takes data outside personal-data rules, subject to your legal advice

Privacy law

Built around the rules you answer to

Hover a page to read between the redactions. Anonymization is a tool that supports your compliance work. It is not legal advice, and your counsel makes the final call.

GDPREuropean Union

1.Faces, plates and voices can be personal data.
2.Share only what you need, and only with a lawful basis.
3.Where possible, anonymise or pseudonymise before wider use.

DPDP ActIndia

1.Digital personal data needs consent or a legitimate use.
2.Keep data only while it serves its stated purpose.
3.Apply reasonable security safeguards throughout.

CCPA / CPRACalifornia, USA

1.Consumers can ask what is collected and shared.
2.Properly de-identified data sits outside many obligations.
3.Limit use of sensitive personal information.

HIPAAUnited States, health

1.Full-face images are among the identifiers to remove.
2.De-identification can follow a defined standard.
3.Access to remaining data stays logged and limited.

Data lifecycle

Your footage is never kept longer than needed

Five steps from upload to verified deletion, each one logged.

UploadEncrypted in transit
TLS transfer or your own secure bucket, with checksums.
ProcessIsolated workspace
Locked-down virtual desktops, no downloads, screen watermarking.
DeliverTo your storage
Anonymized files plus a per-file audit report.
RetainShort, agreed window
Originals kept only as long as you specify, often days.
7d
PurgeVerified deletion
Keys destroyed, data erased, a deletion certificate issued.

Vetted, NDA-bound team
Background-checked reviewers who sign confidentiality terms.
Isolated review desks
No personal devices, no screenshots, no local copies.
Full audit trail
Who saw which file, when, and what they changed.
You hold the keys
Customer-managed encryption for reversible workflows.

Free pilot

One week from sample to proof

Send us a sample. By Friday you have anonymized files, measured accuracy and a fixed-scope quote. Timings are indicative.

MonTueWedThuFri9:0010:0011:0012:0013:0014:00

FAQ

Frequently Asked Questions

Straight answers on image and video anonymization: accuracy, privacy, security, cost and timelines.

It is the process of finding personal identifiers in images and video, such as faces, license plates, screens and readable text, and permanently hiding or replacing them so individuals can no longer be recognised, while keeping the rest of the footage useful.

To share, publish, analyse or train AI on visual data without exposing the people in it, to meet privacy obligations, and to reduce the damage if data is ever leaked.

A managed service: detection tuned to your footage, consistent tracking across frames, your chosen anonymization method, human quality review, secure delivery and an audit report for every batch.

Automotive and mapping teams, retailers and city authorities with camera networks, healthcare and research teams, media organisations, insurers and any company preparing visual data for sharing or AI.

Book a free consultation and share a sample. Within a week we return an anonymized sample, an accuracy report and a fixed-scope quote, so you can judge quality before committing.

REC · 00:00:07Start with a sample

Ready to protect every face in your footage?

Send a short clip or a few hundred images. We will return anonymized files, an accuracy report and a fixed-scope quote within a week, free.

Get a free anonymized sample See the pilot week

See everything. Identify no one.

Six ways to make a person disappear

Gaussian blur

Pixelation

Solid mask

Synthetic replacement

Silhouette & pose

Text & document redaction

If it can identify someone, we find it

Faces

License plates

Signs & addresses

ID badges & cards

Screens & monitors

From raw footage to a clean cut

Secure ingest

Detect

Track

Anonymize

Human QA

Deliver

We choose missing nothing over looking pretty

Every face caught, little noise

Wherever a camera is pointed at people

Automotive & ADAS

Retail & smart stores

Smart cities & CCTV

Healthcare & research

Media & broadcasting

Insurance & legal

Pick how final the anonymization should be

Irreversible anonymization

Built around the rules you answer to

Your footage is never kept longer than needed

Vetted, NDA-bound team

Isolated review desks

Full audit trail

You hold the keys

One week from sample to proof

Frequently Asked Questions

Ready to protect every face in your footage?