01
Never biometric or facial recognition
No biometric, facial or similar recognition is performed that could be directly or indirectly linked to a person.
How it works · Privacy and cybersecurity
Video analytics without personal data, by design
KSI turns your camera streams into anonymous, aggregated metadata inside the GPU: no facial recognition, no recordings, and images never leave the client-controlled environment.
NO BIOMETRIC RECOGNITIONNO VIDEO RECORDINGSDISCARDED IN UNDER 1 MSGDPR COMPLIANT
01 · Introduction and background
A common, mature technology deployed worldwide
Video analytics is a very common activity, widespread globally. Historically it emerged with the mass adoption of security systems in the 1980s and 1990s, with the purpose of extracting useful information about what is happening in video feeds. A basic example is motion detection on a security camera, to raise an alert about a possible intrusion into a private area.
Over time, video analytics has continued to evolve, gaining complexity and use cases, to the point that today it is practically impossible to buy a camera (even for home use) that does not ship with multiple embedded analytics: motion detection, person detection, people counting, heatmaps, and so on.
From 2012, at the origin of modern Artificial Intelligence, when the first neural network for detecting objects in images was invented, numerous efforts have been made to include AI-based video analytics and extract more value from the information available in video streams.
Since it began operating, KSI's technology has been applied across numerous markets in Europe, LATAM and North America, passing numerous commercial and legal evaluations and validations at the companies where it was deployed. Below are the legal and technical elements developed over the years to guarantee the safe use of this technology.
1980-1990
Video analytics emerges with the mass adoption of security systems.
2012
Modern AI is born: the first neural network capable of detecting objects in images.
2018
NVIDIA GPUs with the Turing architecture make AI analytics viable at scale and at a reasonable cost.
2019
KSI VISION begins operating, adding operational and commercial value to the management of spaces open to the public.
TODAY
Commercially and legally validated deployments across Europe, LATAM and North America.
02 · Relevant legal frameworks
Designed not to process personal data
In general, the legal analysis to be carried out concerns the application of personal data laws. To that end, KSI has been designed not to process personal data.
GDPR · The global reference law
Personal data is any information relating to an identified or identifiable living natural person. Different pieces of data which, taken together, can lead to the identification of a particular person also constitute personal data. Data that has been de-identified, encrypted or pseudonymized but can be used to re-identify a person remains personal data and falls within the scope of the GDPR.
Example · Law 5/2014 (Spain)
Personal data laws also have decrees or specific interpretations applied to security systems. In Spain, Law 5/2014 regulates that video recordings may be viewed by security company personnel, and in which cases they may be sent to security and justice organizations for analysis. Its purpose is to prevent recordings from being disseminated and exposing personal data of what happens in them.
KSI is not affected: there are no external agents (such as security personnel) accessing and viewing the videos, and no recording of the events in the analyzed spaces is ever generated.
03 · Privacy by design
How does KSI avoid generating personal data?
KSI avoids holding personal data by applying four design criteria:
02
Only aggregated, anonymized data
Generated data is not reversible, and only the information relevant for commercial use is stored.
03
No recordings or stored images
Processing starts directly from the H.264/H.265 encoded streams and analytics runs in real time. It is impossible for an external agent to access images or video: every analyzed stream is discarded in under 1 millisecond.
04
Parallel, local processing
Processing happens entirely on the server installed on the client's premises, in parallel and independent of the CCTV security system, similar to a DVR with video analytics. Raw images are never transmitted to external servers, and sensitive data remains within a secure, locally controlled environment.
04 · Technical operation
Anonymization guaranteed from the entry point into the GPU
Video analytics equipment generally has a processing unit (CPU) and a graphics processing unit (GPU). Video analytics runs inside the GPU, with the following workflow for anonymization and image processing:
- 01
Decoding wired to the anonymization filter
The compressed video stream (H.264 or H.265) is received and decoded directly by the decoder chip integrated into the GPU. This chip is programmed to send the decoded data immediately to the anonymization filter without exposing the images to any external component. This direct connection ensures there is no possibility of extracting or accessing video frames in their original state.
- 02
Immediate anonymization
Once decoded, the data goes directly to the first layer of the convolutional neural network (CNN), designed specifically to perform anonymization. This layer transforms the visual information irreversibly, discarding any biometric detail or information that could identify a person. Only abstract features required for analysis are retained, such as contours, trajectories or patterns.
- 03
Irreversible transformation
Decoded images cannot be reconstructed or reversed at any later stage. Original images are discarded immediately after anonymization, in under 1 millisecond.
- 04
Metadata processing
After anonymization, the subsequent layers of the neural network process the abstract features to generate metadata such as counts, occupancy densities or movement vectors. This metadata is completely anonymous and contains no visual or biometric information.
- 05
Real-time optimization
This entire flow, from initial decoding to metadata generation, is optimized to run in real time, ensuring every frame is processed in under 1 millisecond.
- 06
Optimized GPU routines
These operations are implemented through routines written in languages such as CUDA, leveraging the GPU's parallel architecture to maximize efficiency and minimize latency.
- 07
Compliance guarantees
This design ensures compliance with privacy standards such as GDPR. There is no storage, transfer or access to the original data at any time, and anonymization is guaranteed from the entry point into the GPU.
05 · Generated data
What data the video analytics generates
The KSI Vision video analytics system is designed to generate only anonymous, aggregated, non-reversible data. It does not produce or store images, video or biometric information: the outputs are structured metadata for operational and business intelligence purposes.
People counting
- What it measures
- The number of people crossing virtual lines or entering and exiting defined zones.
- How it is used
- Monitoring foot traffic, measuring store performance or evaluating space usage.
Format
{
"timestamp": "2025-04-02T10:15:00Z",
"zone": "Entrance A",
"count_in": 12,
"count_out": 8
}Occupancy and density
- What it measures
- How many people are present in a specific area and how they are distributed.
- How it is used
- Detecting crowding, optimizing space layout or complying with safety regulations.
Format
{
"timestamp": "2025-04-02T10:15:00Z",
"zone": "Food Court",
"occupancy": 47,
"density": "medium"
}Heatmaps (spatio-temporal aggregates)
- What it measures
- Aggregated movement patterns and concentration of people over time.
- How it is used
- Identifying hot and cold zones for space planning, signage placement, and more.
Format
{
"zone": "Floor 1",
"heatmap": [[0, 5, 12], [3, 9, 8], ...],
"interval": "15min"
}Dwell time
- What it measures
- How long people remain in a specific zone.
- How it is used
- Evaluating attention levels or waiting times.
Format
{
"timestamp": "2025-04-02T10:15:00Z",
"zone": "Display B",
"avg_dwell_time": 78
}Trajectory tracking (identity-free)
- What it measures
- Anonymous paths taken within a space, based on non-identifiable temporal vectors.
- How it is used
- Understanding circulation patterns, browsing behavior or entry-to-exit journeys.
Format
{
"path_id": "xyz123",
"path": [
{"x": 10, "y": 14, "t": 0},
{"x": 11, "y": 15, "t": 1}, ...
]
}Group detection
- What it measures
- Whether people walk alone or in groups, based on spatial proximity and time.
- How it is used
- Behavioral segmentation or marketing analysis.
Format
{
"timestamp": "2025-04-02T10:15:00Z",
"zone": "Main Hall",
"groups": 3,
"individuals": 5
}Key characteristics of all generated data
Non-personal
No names, IDs, faces or other identifiers.
Anonymous
No link to people or devices.
Aggregated
Data is grouped by time intervals and zones.
Ephemeral
Temporary identifiers (such as trajectories) are deleted immediately after use.
Real-time
All data is generated and transmitted with latency under 1 second.
06 · Conclusion
Privacy by design, validated across jurisdictions
KSI Vision's video analytics technology is designed from the ground up to align with global privacy standards, while providing actionable information for operational and commercial decision-making. By avoiding the capture, storage or processing of any personally identifiable information, and by generating only anonymized, aggregated metadata, KSI guarantees compliance with data protection laws such as GDPR and its regional equivalents.
The system architecture (which processes data entirely inside the GPU in real time, discards video frames immediately after anonymization, and never transmits raw images externally) offers solid technical guarantees against misuse or unauthorized access. This not only protects people's privacy, it also simplifies legal assessments for clients across different jurisdictions.
With proven deployments in Europe, Latin America and North America, KSI Vision has demonstrated that it is possible to combine advanced AI-driven analytics with a privacy-by-design approach. As video analytics continues to expand, solutions like KSI Vision set a new standard for the ethical and secure deployment of computer vision technologies.