Study 05.1
Where should the unit go in the bedroom?
Question
At what height, and at what tilt, does the unit most often see the face of a playing child, while staying out of the child's reach?
Result
13.4%
of the time, the child's face is usable with the unit placed at a lens height of 1.6 m and tilted 20°. On the ceiling (2.3 m), this falls to 7.0%; without cropping on the child, to less than 1%.
Example, as an order of magnitude: if a child plays 3 h a day in the room (an unsourced assumption, to be checked with families), 13.4% represents about 24 minutes of usable face per day, with the future “bedroom” mode (planned, not coded yet). This is not a measurement.
Figure 1
Share of the time the child's face is usable, by unit height
Unit in a corner, tilted 20°. Simulation of a 3.5 × 3.5 m bedroom.
02-modele-3d/etude-position/grille.json, corrected model of 9 November 2026. This is a simulation, not a measurement.See the data
| Height | Usable face | Floor covered (seated child) |
|---|---|---|
| 1.0 m | 19.4% | 99.9% |
| 1.1 m | 18.4% | 99.7% |
| 1.2 m | 17.3% | 99.3% |
| 1.3 m | 16.3% | 98.9% |
| 1.4 m | 15.3% | 98.3% |
| 1.5 m | 14.3% | 97.7% |
| 1.6 m (recommended) | 13.4% | 96.9% |
| 1.7 m | 12.5% | 96.0% |
| 1.8 m | 11.5% | 95.0% |
| 1.9 m | 10.5% | 93.9% |
| 2.0 m | 9.6% | 92.7% |
| 2.1 m | 8.7% | 89.9% |
| 2.2 m | 7.8% | 88.3% |
| 2.3 m (ceiling) | 7.0% | 86.7% |
| 2.4 m | 6.2% | 84.9% |
Correction
7.9% → 13.4%
Corrected on 9 November 2026. The first version kept the seated child's head tilted at 30° permanently: seen from a unit at 1.6 m, the child was almost never facing it. The error was spotted in a review carried out with another AI model, then checked in the code. The unit remains 1.9 times more effective than on the ceiling. See 05.5
Figure 2
Height matters far more than tilt
Share of the time the face is usable, for 285 settings: 15 heights × 19 tilts. The darker the cell, the more often the face is legible.
grille.json, same simulation (9 Nov. 2026). Hover over a cell for its value, click to set the simulator.Figure 3
The simulator
The bedroom in 3D: set the unit, place the child
Choose the height and the tilt, seated or standing, the direction of the gaze; click on the top view to move the child. Clicking a cell in figure 2 also sets the simulator.
The same model as figures 1 and 2, recalculated in the page and checked against the Python calculation over the 285 settings (07-site-prive/outils/tests/test_chambre.mjs). The furniture is decorative: it hides nothing in the model.
And two cameras?
A second camera in the opposite corner doubles the share of the time the face is seen: 26.8% instead of 13.4%. The two cameras never see the same face at the same moment. In two adjacent corners, they overlap a little: 25.7%. It is not planned for now: it would cost about €230 more, complicate synchronisation and strengthen the impression of surveillance.
| Configuration (1.6 m, 20°) | Corrected model | Previous model |
|---|---|---|
| 1 camera in a corner | 13.4% | 7.9% |
| 2 cameras, opposite corners | 26.8% | 15.7% |
| 2 cameras, adjacent corners | 25.7% | 15.0% |
Recalculated on 9 November 2026 with the corrected model (recherches/2026-11-09_etude-deux-cameras-corrigee.json). The earlier study reported 15.9% and 15.1%: the second unit had lost its 9 cm setback from the wall. Once corrected, the previous model gives 15.7% and 15.0%, which does not change the conclusion.
Detail
See the method
Method
A simulation in Python. A child is placed every 10 cm in the bedroom, facing 36 directions, seated or standing, and the share of cases in which the face is usable is counted. This is repeated for 15 heights (from 1.0 to 2.4 m) and 19 tilts (from 0 to 45°).
Assumptions
- Square bedroom of 3.5 × 3.5 m. Unit in a corner, lens 9 cm from the walls, facing the centre of the room.
- Camera Module 3 Wide: 102° horizontally, 67° vertically (official datasheet).
- Cropping on the child: the fineness of the 12 MP sensor is kept (4,608 px across 102°).
- Child aged 2 to 7, always more than 30 cm from the walls: seated on the floor 60% of the time (face at 0.55 m, head tilted by 18° on average, standard deviation 14°), standing 40% of the time (face at 0.95 m, head tilted by 10°).
- Usable face: within the field of view, seen at less than 45° from front-on, and at least 48 px wide (for a 12 cm face).
Figures
| Setting | Usable face | Floor covered | Blind zone at the foot of the corner |
|---|---|---|---|
| 1.6 m, 20° (recommended) | 13.4% | 96.9% | 0.78 m |
| 1.6 m, 37.5° (best setting, out of reach) | 13.4% | 99.9% | 0.36 m |
| 1.7 m, 20° | 12.5% | 96.0% | 0.85 m |
| Ceiling, 2.3 m, 20° | 7.0% | 86.7% | 1.29 m |
| 1.0 m, 25° (within a child's reach) | 19.4% | 100% | 0.28 m |
| 1.7 m, 25°, no cropping | 1.0% | 97.7% | 0.70 m |
| 1.6 m, 20°, previous model (head fixed at 30°) | 7.9% | — | — |
Source: recherches/2026-11-09_etude-placement-resultats-corrigee.json. “Floor covered”: the share of the room where a seated child is within the field of view. “Blind zone”: the distance from the corner within which a seated child falls outside the field of view.
Why 20° rather than 37.5°
At 1.6 m, the tilt changes the result very little between 7.5° and 40°. 20° is kept, the angle of the design, which sees the body of a seated child from 0.78 m from the corner. The 37.5° setting reduces the blind zone to 0.36 m: this is a lead for V2.
Limits
- A simplified model, not a measurement: no furniture, no body hiding the face, a single typical bedroom.
- Head posture (18° on average) is an assumption. It will be checked with the real unit, after assembly.
- The 48 px threshold is a working threshold adopted for the model, with no published reference.
- The figure says how long the face is legible, not what the analysis makes of it.
Files
02-modele-3d/etude-position/etude_position.py, grille.json, etude_deux_cameras.py; design journal, steps 8, 10 and 14.
