No RGB capture
Depth geometry can reduce identification risk, but the full data path still needs governance.
Privacy-Safe Fall Detection With Depth, Not Video
Build privacy-first fall detection for bathrooms and bedrooms using depth-only 3D geometry, edge processing and application-specific validation.
Plan My EvaluationWhy depth geometry
A privacy-aware starting point for room-level fall detection
3D ToF generates geometric distance data rather than RGB video. The host can analyze posture, floor boundaries and room context in real time while reducing the privacy surface in bathrooms and bedrooms.
Works without visible light
Active near-infrared sensing does not depend on a lit room. Materials and optical conditions still matter.
Edge-ready data path
Keep raw frames local and send only the event fields the product actually needs.
Where Dense 3D Geometry Matters Most
Qualify the room before choosing a module. The useful signal is the relationship between a person, the floor, the bed and the surrounding furniture.
How the room is interpreted
Bathrooms and toilets
Assess slips, slow collapses, long low-position events and recovery near the toilet, shower or wet floor.
Bedrooms and bedside zones
Separate bed height, floor height, bedside sliding and fast movement onto a mattress with room geometry.
Care rooms with intervention
Include walkers, carers, partial occlusion and more than one person in the data and acceptance plan.
Compare another sensing mode when persistent steam, no direct line of sight, through-wall sensing or vital signs are central requirements.
From Active Light to an Actionable Event
The module is one layer in the product. Make the sensing, perception and data handoffs explicit so engineering and privacy reviews can follow the evidence.
Active near-infrared light
The emitter sends modulated light into the scene.
Per-pixel time of flight
The sensor estimates distance for each valid pixel.
Depth and confidence
The frame carries depth, signal quality and invalid flags.
Geometry and posture
The host can derive point clouds, planes and posture features.
Fall-state decision
Application logic evaluates a sequence, not one low-height frame.
Raw depth
Use locally for segmentation, calibration and scene validation. Default path: no upstream storage.
Derived features
Use posture, plane distance, trajectory and confidence features where the application needs them.
Event data
Send only the room or device ID, time, event type and confidence required by the customer system.
No RGB does not mean no data governance. Define whether IR intensity, raw depth, derived pose and event data are enabled, stored, transmitted and accessible.
Architecture + wavelength
Choose the Technology From the Failure Mode
Area-array iToF is usually the stronger first evaluation for dense posture and floor geometry. Compare SPAD dToF, 850 nm or another sensing mode when scattering, range, steam or hidden mounting dominates.
| Route | What it observes | Qualification question |
|---|---|---|
| 3D ToF depth | Dense depth map or point cloud for body, floor, bed and furniture geometry. | Strong starting point for posture boundaries and explainable room geometry. Validate steam, mirrors, MPI and occlusion. |
| 60 or 77 GHz mmWave | Range, velocity and angle data with motion and micro-motion potential. | Compare when persistent steam, optical occlusion or hidden mounting is central. Validate multipath and ghost targets. |
| Thermal imaging | Two-dimensional temperature distribution and human heat silhouette. | Consider when temperature contrast and cost fit the project. Validate reflections, ambient heat and occlusion. |
| PIR or sensor fusion | Low-detail occupancy trigger, or complementary coverage for independent failure modes. | Useful as a wake-up or fusion input, but test slow collapses, pets, HVAC movement and room-level blind spots. |
940 nm: near-range and quiet-room starting point
A practical first comparison for bedrooms, bedside zones and near-range bathrooms. Verify ambient-light margin, detector response and the complete IEC 60825-1 Class 1 system path.
850 nm: reach when geometry requires it
Consider it when ceiling mounting needs longer reach or wider coverage. Treat visible glow, resident comfort, ambient response and eye-safety margins as acceptance criteria.
The optical-axis pitch angle, diagonal distance, pixel coverage, emitter drive, cover window and host compute budget should be evaluated as one design. No iToF or dToF label removes the need for scene testing.
Validate the Failure Modes, Not Just the Sensor
A credible fall event combines motion, posture, support planes, duration, confidence and recovery. Steam, condensation, wet windows, mirrors and glossy tiles can also create ambiguous depth.
Bathroom test sequence
-
Dry baseline
Record valid-depth coverage, range bias and track continuity before water or steam is introduced.
-
Shower rise
Measure temperature, humidity and optical-window conditions as the room becomes wet and warm.
-
High humidity
Check confidence, invalid pixels, false clusters and behavior across the target mounting angle.
-
Condensation recovery
Record wet-window behavior after water stops and define a visible degraded mode.
Materials and reflections to include
Dry and wet ceramic tile
Mirror and glass shower door
Black mat, white towel and curtain
Water film, cleaner residue and fingerprints
Toilet, grab rail and metal fixtures
Monitor valid-depth coverage, distance bias, signal strength, confidence, false clusters, track interruptions and recovery time. Define degraded mode instead of silently accepting bad output.
A fall is a sequence
Stable standing or movement
Clear downward motion
Near floor or unusual support plane
Sustained low position
No autonomous recovery or abnormal recovery
Negative samples are part of the product
Fast movement onto a bed
Sitting on the bed edge or floor
Bending to pick up an object
Squatting, kneeling, praying or stretching
Slowly standing from the toilet
Carer assistance and partial occlusion
Datasheet-based shortlist
Recommended Module Selection Matrix
Start with room range, field of view, frame rate and host interface. These candidates support evaluation; the final choice still depends on enclosure, condensation, EMC and acceptance testing.
Compact SPAD dToF path
DMAS2M001
- 40 x 30 at 10 fps
- 0.2-8 m @ 30 klux
- 60 x 45 degrees
- USB-C / FPC
Compare when time gating, scattering or range separation dominates.
View product details
Compact embedded path
DMOM2508CL
- 320 x 240, 10-30 fps
- 940 nm: 0.2-2 m
- 71.8 x 56.6 degrees
- MIPI CSI-2
Start here for near-range or lower wall mounting.
View product details
Mid-range embedded path
DMOM2808D
- 320 x 240, 10-30 fps
- 850 nm: 0.25-4.5 m
- 71.8 x 56.5 degrees
- MIPI CSI-2
Consider for ceiling mounting and larger bedrooms.
View product details
Low-data wide field of view
DMOS5031A
- 100 x 100, up to 20 fps
- 940 nm: 0.2-2.5 m
- 60 x 60 degrees
- UART or SPI
Consider when lower bandwidth supports data minimization.
View product detailsIP rating, condensation, disinfectant exposure, optical window and EMC remain customer-system validation items.
Questions to Resolve Before Room Testing
Direct answers to the privacy, environment, algorithm and module questions that shape an evaluation brief.
Plan My EvaluationIs 3D ToF fall detection privacy-safe?
Depth-only 3D ToF does not capture RGB color or facial texture, which materially reduces identification risk. Privacy still depends on the complete system: process frames locally, transmit only necessary event data, restrict raw-depth retention, control access and assess whether the remaining data can be linked to an individual.
When is 3D ToF better than 60 GHz mmWave for fall detection?
3D ToF is often the better starting point when the algorithm needs dense body geometry, floor distance, posture boundaries and room context. mmWave is often stronger under persistent steam, optical occlusion or when micro-motion is required. The correct choice depends on mounting, materials, privacy architecture and validation data.
Will 3D ToF work in a dark, steamy bathroom?
Visible darkness is not the main limitation because ToF uses active near-infrared illumination. Steam, condensation, wet optical windows, glossy tiles and mirrors can still reduce signal or produce invalid and multipath depth. Qualification must measure valid-depth coverage, range bias, false clusters and recovery after condensation.
Can a ToF sensor distinguish bending down from a fall?
It can provide useful features for that distinction, including centroid height and velocity, body axis, support-plane distance, contact area, trajectory continuity, low-position duration and recovery. The customer or algorithm partner still needs scene-specific data and acceptance criteria.
Can it distinguish falling onto a bed from falling onto the floor?
A depth system can model bed height, floor height and the body relationship to each support plane. Bed geometry, occlusion, bedding and mounting angle still require direct validation, especially for fast movement onto a mattress.
Does DOMI provide a complete fall-detection algorithm?
DOMI provides depth modules, data outputs, SDK and reference integration support. The final fall classifier, data set, thresholds, alert policy and product validation remain with the customer or its algorithm partner unless a separate scope defines otherwise.
Which DOMI module should be used for a ceiling-mounted design?
DMOM2808D is a candidate for mid-range embedded evaluation with a published 0.25-4.5 m range and 71.8 x 56.5 degree field of view. The correct choice still depends on the actual diagonal distance, pixel coverage, wavelength, ambient light, enclosure and validation results.
Is the module waterproof or IP-rated?
Do not infer an IP or condensation rating from the module name. The enclosure, optical window, seals, cleaning process, thermal design and final EMC and ingress tests belong to the customer system qualification.
Does a depth-only system require a HIPAA BAA?
BAA applicability depends on whether DOMI or its service processes protected health information on behalf of a regulated entity. A hardware supplier that does not receive or maintain that data should not claim BAA status by default. Review the actual data flow and contract roles.
Can raw depth frames remain entirely on-device?
The architecture can be designed so raw depth is processed locally and not sent upstream. Confirm whether IR intensity, diagnostic frames, derived pose or event history are also enabled, stored or exposed by the final product.
What test data should be collected before production selection?
Collect representative room geometry, mounting positions, lighting, steam, condensation, mirrors, wet surfaces, occlusion, daily actions, fall types, people variation, network faults and recovery behavior. Report recall, precision, false alarms, latency and sensor-health indicators by condition.
Plan your evaluation
Define Your Fall-Detection Evaluation With a DOMI FAE
Share your room geometry, operating conditions and integration path. We will help shortlist the sensor, identify validation risks and define the next engineering test.
- Application-fit review for ToF, radar or a fused route
- EVK and module shortlist by range, FoV, rate and interface
- Preliminary mounting and blind-zone review
- Bathroom validation checklist for steam, mirrors and wet surfaces
- Depth data, SDK and host-platform availability check
- NDA path before detailed file exchange
Send your engineering brief
A few details are enough to start the conversation. Add dimensions, mounting height or host-platform details if they are available.