Video anonymization software definition
Video anonymization software is designed to detect and mask personal data visible in images and video footage. In practice, for visual materials, this mainly means faces and license plates. The goal is to reduce the possibility of identifying a natural person before the material is used further, for example through publication, sharing, archiving, analysis, or transfer to another party.
Under the GDPR approach, anonymization means processing data in such a way that a person can no longer be identified using means that are reasonably likely to be used. This follows from Recital 26 of Regulation (EU) 2016/679. In practice, for video materials, some solutions marketed as “anonymization” actually provide permanent image masking rather than anonymization in the strict legal sense. That is why effectiveness depends on whether a person can still be re-identified after masking based on a face, license plate, or other contextual elements.
This type of software typically uses computer vision models for object detection and then applies a mask, blur, pixelation, or area fill. For face anonymization and license plate blurring, automatic performance depends on the quality of the model, footage resolution, camera angle, lighting, motion, compression, and partial occlusion of the object.
In the context of Gallio PRO, the term refers to software that automatically detects and blurs faces and license plates in videos and photos. It does not include automatic detection of logos, tattoos, name badges, documents, or content displayed on monitors. These elements can be masked manually in the editor. The software does not perform real-time anonymization or live video stream anonymization.
How video anonymization software works
Technically, the process consists of several stages. First, the system analyzes video frames and locates objects that require protection. It then tracks them from frame to frame and applies a mask throughout the recording. This reduces the risk of mask “flickering” or a face being briefly exposed.
Automatic face anonymization and license plate anonymization usually rely on deep learning models. These models are widely used to build detectors that can handle image variability, such as different face positions, partial occlusion, vehicle movement, or changing weather conditions. The model is trained in advance on labeled datasets and then used for inference, meaning object detection in new materials.
- detection - identifying the face or license plate area in a single frame,
- tracking - maintaining identification of the same object across consecutive frames,
- masking - applying blur, pixelation, or full concealment,
- verification - operator quality control, especially for high-risk materials.
In practice, a good video anonymization solution should combine automation with manual correction. Detection alone does not eliminate errors such as false negatives, meaning missed faces or license plates. From the perspective of a Data Protection Officer, this is a critical parameter, because even a single omission may result in the disclosure of personal data.
Types of video anonymization software
When choosing video anonymization software, the deployment architecture and the degree of automation matter. These features affect legal risk, data security, operating costs, and process efficiency.
Criterion | On-premise | Cloud
|
|---|---|---|
Processing location | Own or dedicated infrastructure | Provider environment |
Control over data | High | Depends on the contract and configuration |
File transfer outside the organization | May not be required | Usually required |
Infrastructure requirements | Higher | Lower on the user side |
Assessment for sensitive materials | Often preferred | Requires analysis of transfers and processors |
The second distinction concerns the operator workflow. Automatic solutions increase processing scale but must be supervised. Manual solutions are slower, but they allow masking of objects not detected by the model.
Operating mode | Use case | Limitations
|
|---|---|---|
Automatic | Large volumes of footage, repetitive scenes | Risk of omissions and detection errors |
Manual | Non-standard materials, objects outside the model scope | More time-intensive |
Hybrid | Most common operating model | Requires a quality control procedure |
Key parameters and evaluation metrics
Software assessment should not rely only on the label “AI.” Measurable parameters are needed. In object detection systems, precision, recall, and related metrics are the most common. These definitions are standard in machine learning and the computer vision literature.
- precision = TP / (TP + FP) - what percentage of detections was correct,
- recall = TP / (TP + FN) - what percentage of actual objects was detected,
- F1 = 2 × (precision × recall) / (precision + recall),
- IoU - Intersection over Union, a measure of how well the detection box matches the object,
- processing speed - for example, frames per second or time per 1 hour of footage,
- tracking stability - the number of masking interruptions in a sequence,
- critical error rate - the number of missed faces or license plates per unit of recording time.
For compliance use cases, high recall is usually more important than precision alone. False positives increase the number of unnecessary masks, but false negatives can lead to personal data disclosure. For this reason, testing should be performed on the organization’s own reference materials, not only on the vendor’s data.
Criteria for choosing video anonymization software
The choice of solution should result from a risk analysis, the type of material, and the legal basis for processing. For a Data Protection Officer, both technical features and organizational safeguards are important.
- scope of automatic detection - whether the system detects faces and license plates,
- manual correction capability - essential for objects not automatically supported,
- deployment model - on-premise or cloud,
- absence of unnecessary logs containing personal data,
- support for photos and videos in the required formats and resolutions,
- performance with large collections of material,
- ability to validate the result before exporting the output file,
- technical documentation and the ability to audit process settings.
In the case of Gallio PRO, an important point is that the software does not store logs containing face and license plate detection results or other logs containing personal data or special category data. From a data minimization perspective, this is an important organizational feature.
Legal and regulatory aspects
The main point of reference remains the GDPR, namely Regulation (EU) 2016/679 of 27 April 2016. The principles of privacy by design and privacy by default under Article 25, as well as security of processing under Article 32, are also relevant. Recital 26 sets out the criterion for assessing identifiability after anonymization.
With regard to faces, the obligation to apply face blurring may also arise from rules on image rights and personal rights, including the Civil Code and the Copyright and Related Rights Act. In practice, three typical exceptions are indicated where no permission is required to publish a person’s image: a widely known person in connection with the performance of public functions, an image constituting only a detail of a larger whole such as an assembly, landscape, or public event, and a situation where the person received agreed remuneration for posing.
For license plates, the legal assessment is not uniform. In European countries, license plate blurring is often treated as a safe data protection practice, but the legal classification depends on the context and national law. In Poland, there is divergence. Guidance from data protection authorities and EU case law supports a cautious approach, while some administrative court rulings assume that a registration number does not always constitute personal data. From a compliance perspective, this means that the context and the risk of identification must be assessed.
Limitations and practical use cases
No video anonymization software can guarantee flawless performance for every material. Problems arise with very low resolution, heavy compression, fast camera movement, large crowds, night scenes, and partial object occlusion. For this reason, the process should include quality control, especially before publishing material online.
Video anonymization software is used by local government units, police forces, road authorities, transport companies, media organizations, surveillance operators, and entities maintaining technical field documentation, among others. In each of these cases, the goal is similar: to enable further use of the footage without unnecessarily disclosing the identity of individuals and vehicle registration numbers.