Element 13.20
Cameras and machine vision
- Sheet
- 13.20
- Chapter
- 13 Digital infrastructure
- State
- Drafted
- Release
- 16 November 2026
- Editor
- Contributed by Praecise as open research
- Reviewers
- To be named at review
- Safety
- S1, as the chapter
- Licence
- CC BY-SA 4.0
- Resources
- 10
The element
- Requirement
Cameras and other imaging sensors are chosen for the task they serve, process images locally where possible, and are run within the privacy law of the place. People know where cameras watch, footage is kept only as long as needed, and identification of people by their faces or bodies is not done without a lawful basis.
- How to tell
- The camera matches the motion and light. A global shutter captures every pixel at once and suits fast motion; one documented module uses 3.45 micrometre pixels at 1.6 MP for this (Raspberry Pi camera documentation). Larger sensors matter more than resolution at night, says one video-analytics project (Frigate recommended hardware).
- Depth is measured where a machine must judge distance. One stereo camera states depth error below 2 percent under 4 m and below 6 percent at 7 to 10 m (Luxonis OAK-D Pro).
- Detection runs on local hardware. A local video recorder documents inference of about 5 to 40 ms per frame on accelerators from several makers and advises CPU-only detection for testing only (Frigate recommended hardware, Frigate).
- Where European law applies: a camera covering even part of a public space falls outside the household exemption; a warning sign at about eye level gives the purpose, the controller and the rights before a person enters; footage is in most cases erased after a few days, and storage beyond 72 hours needs more justification (Guidelines 3/2019 on processing of personal data through video devices). Biometric data used to identify a person uniquely is prohibited unless an exception applies (General Data Protection Regulation, Article 9: special categories of personal data).
- Options
- Camera modules for single-board computers: 12 MP autofocus, high-quality with interchangeable lenses, global shutter, and one with on-sensor inference (Raspberry Pi camera documentation), driven by an open camera stack (libcamera).
- Stereo depth cameras with on-board AI, active infrared for darkness (Luxonis OAK-D Pro).
- Non-camera ranging that sees no faces: a spinning 2D lidar with 0.15 to 12 m range (Slamtec RPLIDAR A1); 60 to 64 GHz radar with reference designs for people counting and presence (Texas Instruments IWR6843 mmWave radar sensor).
- Local video analytics with object detection and Home Assistant integration (Frigate), using open model families for detection, segmentation and pose (Ultralytics YOLO models).
- Thermal and multispectral imaging: no maker or evaluation page was opened in this work; see the research note.
- Settled, open, contested
Settled: local detection on a small accelerator is practical (Frigate recommended hardware). Settled in European law: outward-facing recording leaves the household exemption (Guidelines 3/2019 on processing of personal data through video devices). Open: accuracy of detection models in the community's own scenes; test before relying on them. Contested: cameras for safety in shared spaces. Some hold they deter harm and help find lost people; others that they change how people behave and erode trust. Radar or lidar presence sensing is a middle way where only counts are needed (Texas Instruments IWR6843 mmWave radar sensor).
The four stages
- Learn: Raspberry Pi camera documentation; the European guidelines (Guidelines 3/2019 on processing of personal data through video devices).
- Design: choose shutter, sensor size and lens; decide what must not be recorded (General Data Protection Regulation, Article 9: special categories of personal data).
- Build: libcamera, Luxonis OAK-D Pro, Frigate recommended hardware.
- Deploy: Frigate with signage, retention and access rules written down.
Resources at each stage
1 Learn
Understand
2 Design
Size and choose
- Guidelines 3/2019 on processing of personal data through video devices Guide, S1
- Frigate recommended hardware Guide, S1
- General Data Protection Regulation, Article 9: special categories of personal data Standard, S1
- Luxonis OAK-D Pro Reference design, S0
- Raspberry Pi camera documentation Reference design, S0
- Slamtec RPLIDAR A1 Reference design, S0
- Texas Instruments IWR6843 mmWave radar sensor Reference design, S0
3 Build
Build or obtain
- Frigate recommended hardware Guide, S1
- Frigate Tool, S1
- libcamera Library, S0
- Luxonis OAK-D Pro Reference design, S0
- Raspberry Pi camera documentation Reference design, S0
- Slamtec RPLIDAR A1 Reference design, S0
- Texas Instruments IWR6843 mmWave radar sensor Reference design, S0
- Ultralytics YOLO models Library, S0
4 Deploy
Install, run and measure
What changes between 30 adults and 300 people
At 30 adults, a few cameras for wildlife, a workshop or a robot, with no recording of shared spaces, need little governance. At 300 people, any camera on paths, gates or shared rooms needs a written purpose, signage, a retention period, a named controller and a record of who may view footage.
Reference designs
-
Reference design
Maker's documentation for a stereo depth camera with on-board AI: two global-shutter mono cameras on a 7.5 cm baseline, a colour camera, an infrared dot projector and flood light, stated depth error below 2 percent under 4 m and below 6 percent at 7 to 10 m, 4 TOPS of on-board compute and up to 15 W draw.
External resource. Not made or verified by Public ProtocolMaturity: EstablishedSafety S0CommercialListed 16 November 2026 -
Reference design
Raspberry Pi camera documentation
Maker's documentation for its camera modules, listing sensor, resolution and pixel size: Camera Module 3 (12 MP, autofocus, HDR mode, NoIR option), the High Quality Camera (12 MP, interchangeable lenses), the AI Camera (on-sensor inference) and the Global Shutter Camera (1.6 MP, 3.45 micrometre pixels, all pixels captured at once for fast motion).
External resource. Not made or verified by Public ProtocolMaturity: EstablishedSafety S0CommercialListed 16 November 2026 -
Reference design
Maker's specification for a low-cost spinning 2D laser scanner used on small mobile robots: 0.15 to 12 m range, 1 degree or finer angular resolution, 8,000 samples per second, 5.5 Hz scan, 0.5 W, rated 0 to 40 degrees C, USB connection.
External resource. Not made or verified by Public ProtocolMaturity: EstablishedSafety S0CommercialListed 16 November 2026 -
Reference design
Texas Instruments IWR6843 mmWave radar sensor
Maker's page for a single-chip 60 to 64 GHz FMCW radar with on-chip DSP and hardware accelerator, rated -40 to 105 degrees C junction temperature, with reference designs for people counting, presence and gesture sensing and automated doors.
External resource. Not made or verified by Public ProtocolMaturity: EstablishedSafety S0CommercialListed 16 November 2026
Parts list
Every resource that supports this element. Those marked External carry the label “External resource. Not made or verified by Public Protocol”.
| Resource | Type | Origin | Maturity | Safety | Licence, as stated at its source |
|---|---|---|---|---|---|
| Guidelines 3/2019 on processing of personal data through video devices | Guide | External | Established | S1 | Not stated on the page opened |
| Frigate recommended hardware | Guide | External | Established | S1 | Not stated on the page opened |
| Frigate | Tool | External | Established | S1 | MIT (name and logo are trademarks, not covered) |
| General Data Protection Regulation, Article 9: special categories of personal data | Standard | External | Established | S1 | Not stated on the page opened |
| libcamera | Library | External | Established | S0 | Not stated on the page opened |
| Luxonis OAK-D Pro | Reference design | External | Established | S0 | Not stated on the page opened |
| Raspberry Pi camera documentation | Reference design | External | Established | S0 | Not stated on the page opened |
| Slamtec RPLIDAR A1 | Reference design | External | Established | S0 | Not stated on the page opened |
| Texas Instruments IWR6843 mmWave radar sensor | Reference design | External | Established | S0 | Not stated on the page opened |
| Ultralytics YOLO models | Library | External | Established | S0 | Not stated on the page opened |
Depends on, and completes
It is one of 25 elements that complete Digital infrastructure.
The standing questions
Who may watch, and who watches the watchers. Who decides where a camera goes. Who can ask for footage to be deleted.
Environments
Darkness
active infrared projection and flood light give depth and vision at night (Luxonis OAK-D Pro).
one low-cost lidar is rated only 0 to 40 degrees C (Slamtec RPLIDAR A1); one radar chip is rated -40 to 105 degrees C junction temperature (Texas Instruments IWR6843 mmWave radar sensor).
- Islands and coasts, tropical humid forest, temperate forest, high altitude, flood plains: no source found.
Environment profiles for these settings: Arctic, subarctic and permafrost, including sea ice; Extreme cold; Hot deserts; Extreme heat. All environments
Safety
State and evidence
Drafted
An editor has written the element's text from the resources listed under it, citing them. It is checked when it is reviewed.
Evidence from a real place or system is recorded here, with what did not work. What the states mean
History
- Drafted in the release of 16 November 2026.