UWB vs 60 GHz Radar vs Cameras for Automotive Child Presence Detection

Choosing a child presence detection sensor looks like a datasheet comparison and is not one. Three technologies are competing for the same slot in the cabin, every vendor publishes numbers that flatter their own silicon, and two of them make claims about each other that cannot both be true. Meanwhile the thing that actually decides the outcome is a test protocol with a six year old dummy, a 300 GSM cotton blanket and a footwell.

This article compares ultra wideband radar, 60 GHz millimetre wave radar and camera based in-cabin sensing against the bar that matters, which is the Euro NCAP Child Presence Detection protocol rather than any vendor’s benchmark. It sets out what each technology does well, where each one breaks, what the two radar camps say about each other, and the specific question that resolves the argument for a given programme.

One disclosure up front, because it shapes how you should read the recommendations. needCode builds in-cabin sensing on UWB radar and is a certified Qorvo partner. The conclusion below is still that 60 GHz wins the standalone case, and the reason that conclusion is worth reading is that it goes against our own product line.

The bar all three have to clear

Before comparing sensors it is worth being precise about what the protocol demands, because two of the three technologies fail on requirements that have nothing to do with detection range.

Euro NCAP requires direct sensing. The system must detect a sign of life, meaning movement, respiration or heartbeat. From 2025 onward, indirect systems that infer a child from a door opening sequence or a seat pressure mat score zero, no matter how well they work. That single rule removed an entire product category from the market.

The test conditions are specific and unkind. Four dummy configurations are used, from a 4 kg newborn in a rearward facing carrier through to a Q6 six year old on a booster cushion. Children are covered with a blanket of at least 70 by 90 cm at 300 GSM, arranged either shoulders to feet for a sleeping child or chest down for an awake one. Rearward facing seats get a sun shade attached to the carry handle. A system that reads a clear chest in good light is not being tested.

Coverage is where the geometry bites. Scenarios 1 and 2, the forgotten child and the intentionally left child, cover all rear seating positions across every seat adjustment. Scenario 3, where a child climbs into an unlocked vehicle on their own, extends coverage to the driver’s seat and to all footwell areas and anywhere else in the cabin a child might hide. A child presence detection sensor that cannot see into a footwell caps out at the lower scoring tiers.

The Euro NCAP CPD test and assessment protocol is the document to read rather than any summary of it, including this one. The respiration benchmarks alone are worth the download: 30 breaths per minute for a newborn falling to 18 for a six year old, which is the sensitivity floor your signal chain has to clear.

Timing is a separate constraint that affects power budget more than sensor choice. The initial warning fires within 15 seconds of locking, escalation begins within 90 seconds of that warning ending and repeats at least every minute for 20 minutes, and intervention is required within 10 minutes of locking. Your child presence detection sensor is running on a parked vehicle’s battery for that whole window.

60 GHz millimetre wave radar

This is the incumbent and, for a standalone installation, the one to beat.

60 GHz displaced 24 GHz for in-cabin work because bandwidth buys resolution. Operating in the 57 to 64 GHz region with up to around 5.5 GHz of bandwidth, a 60 GHz sensor reaches roughly 5 cm range resolution, which is enough to separate an adult from a child and enough to resolve the sub-millimetre chest displacement of breathing. 77 to 81 GHz would resolve better still but that band is occupied by exterior ADAS radar and brings a different cost structure.

The coverage argument is the strong one. Texas Instruments published test results for meeting Euro NCAP CPD requirements in which a single overhead mounted AWRL6432 detected a breathing doll across two rows of seats, in both rear facing and forward facing child seats, and lying in the footwell of the front and second rows. One sensor, two rows, footwells included. That maps directly onto the four point scoring tier. Average power was under 10 mW over a 500 ms frame period, which survives the 15 minute monitoring requirement comfortably.

Cost follows from sensor count. Full cabin coverage from one module lands around 35 dollars in the comparisons published by radar vendors, and the same module can be amortised across seat occupancy classification, gesture recognition and intrusion detection rather than sitting idle between CPD events.

The weakness is that it is an additional component. A 60 GHz child presence detection sensor is a new part number, a new antenna aperture in the headliner, a new homologation activity and a new supplier relationship, none of which existed in the vehicle before.

UWB radar

UWB approaches the same problem from the opposite direction. Rather than a purpose built sensor, it reuses a radio the vehicle may already have.

The physics works. A UWB radar transmits short pulses across more than 500 MHz of bandwidth, which gives sub-centimetre range resolution, and the channel impulse response it recovers contains the periodic displacement of a breathing chest. Qorvo’s technical description of UWB radar sensing puts the detection floor at sub-centimetre micro-movements associated with breathing and heart rate, notes that the signal penetrates blankets, bedding and clothing, and quotes under 10 mW for presence detection at several metres. Ceva’s in-cabin vital signs white paper goes further with a processing chain that separates respiration at 0.06 to 0.7 Hz from cardiac activity at 0.75 to 2 Hz, reports better than 1 breath per minute resolution against the Euro NCAP newborn range, and uses blind source separation to pull apart multiple occupants.

The commercial argument is hardware reuse. A vehicle shipping CCC Digital Key already carries UWB anchors, antennas and RF chains for secure ranging. Turning some of that into a child presence detection sensor is, in BOM terms, close to free, and the radio is already homologated.

That argument holds, but it is softer than it sounds, and this is where most evaluations go wrong. Digital key anchors are placed to range accurately to a phone outside the vehicle. Their apertures point outward, their geometry is optimised for the approach zones around the doors, and nothing about that placement was chosen to illuminate a second row footwell. Zero incremental BOM cost is not zero incremental engineering cost. You are either accepting compromised in-cabin coverage from anchors positioned for something else, or adding an interior anchor, at which point you have bought a sensor after all.

Cameras

Cameras are in this comparison because every vehicle building a driver monitoring system already has one, and the temptation to make it earn its keep is strong.

What a camera does well is classification. It tells you a child seat is fitted, which way it faces, whether a seat is occupied by a person or a bag, and increasingly it estimates posture and gaze. For the occupant monitoring requirements arriving alongside CPD, that is genuinely valuable and radar cannot replace it.

What a camera cannot reliably do is clear the CPD bar on its own. Two failures are structural rather than solvable with a better model. The first is occlusion: vision based vital sign extraction needs to see skin or at least a moving surface, and the protocol deliberately covers the child with a blanket, adds a sun shade to the rearward facing carrier, and turns the seat away from the lens. The second is the footwell. A headliner mounted camera with a view of the seating surfaces has no line of sight into the footwell of its own row, let alone the row behind, and scenario 3 requires exactly that.

This is not a minority view among vision vendors. Anyverse, who sell synthetic training data for in-cabin perception, state plainly in their Euro NCAP 2026 readiness guidance that relying solely on RGB will be insufficient for safety critical edge cases and that the 2026 requirements demand sensor fusion with radar, infrared or depth. When the people selling camera training data tell you cameras are not sufficient alone, the question is settled.

The honest position is that a camera is not a child presence detection sensor. It is a classification layer that makes a radar based system smarter and reduces false positives, and it should be specified as that.

The head to head

The table below sets the three candidates against the criteria that decide a child presence detection sensor selection, in the order a programme usually hits them.

Two rows in that table deserve expansion.

Privacy is underweighted in most sensor selections and it is not a soft factor. A CPD system monitors a locked, parked vehicle for 20 minutes after the driver has walked away. Doing that with a camera means an imager is recording the interior of a private vehicle while nobody is present to consent, which is a data protection conversation the radar options simply do not have. Both radar technologies detect a living presence without capturing an image.

Sensors needed for full cabin is where the two radar camps openly contradict each other, and it is the single most consequential disagreement in this comparison.

The vendor contradiction you will run into

Novelic, who manufacture 60 GHz FMCW solutions, published a comparison of 60 GHz FMCW radar against UWB for in-cabin monitoring which states that UWB cannot monitor more than one seat row, that six anchors are needed for coverage, that the resulting system costs around 100 dollars against 35 for their own, and that UWB is unsuited to cabin gesture recognition. Their conclusion is that 60 GHz is the optimal automotive solution.

Qorvo and Ceva, who sell UWB silicon and the DSP IP that runs on it, describe UWB radar detecting newborn respiration through blankets at under 10 mW with sub-centimetre resolution and effectively zero incremental BOM cost when digital key hardware is reused.

Both are technically literate. Both have a commercial interest. They cannot both be describing the same system.

The way to resolve it is to stop reading comparisons, including this one, and change the question you ask vendors. Not “can your child presence detection sensor detect breathing”, because every vendor will say yes and every vendor will be telling the truth under some conditions. Ask instead:

Show me detection data for a Q6 dummy in the second row footwell, under a 300 GSM blanket, from the anchor or sensor positions my vehicle will actually use. Ask for the raw channel impulse response or range-Doppler output, not a pass or fail. Ask what the detection margin was, not whether it detected.

That question collapses the disagreement, because it removes the vendor’s freedom to choose favourable geometry. Most of the gap between the two camps comes down to where the sensor is mounted rather than to the underlying physics, and every published comparison quietly assumes its own best case placement.

Regulatory position

The two radar options for a child presence detection sensor sit under entirely different regimes, which affects homologation effort and sometimes the final design more than the sensing performance does.

The RF exposure detail catches teams out. Both regimes permit up to 40 dBm EIRP in the millimetre wave band, but an in-cabin sensor sits close to occupants, and the power density at the nearest occupant must stay below 1 W/m². In practice that exposure limit, not the EIRP ceiling, is what constrains a 60 GHz in-cabin design. Budget for the assessment early rather than discovering it during homologation.

For UWB there is a subtler issue. The in-cabin sensing function shares its spectral density budget with the digital key ranging function, so a design that adds sensing to an existing anchor is not only sharing antennas, it is sharing an allowance. The two functions have to be scheduled against each other.

What actually decides it

Strip out the vendor noise and the decision reduces to a single question about your vehicle rather than a judgement on the sensors.

If the vehicle does not ship UWB for a digital key, specify 60 GHz. A single overhead mmWave module has published evidence of covering two rows including footwells, it hits the four point scoring tier, it costs around 35 dollars, and it carries no dependency on anything else in the vehicle. Choosing a UWB child presence detection sensor here means paying for UWB hardware you otherwise would not have, for a sensing job another technology does better from one mounting point.

If the vehicle already ships UWB digital key, evaluate UWB seriously, then verify the geometry. The BOM argument is genuine and the physics is sound. What has to be proven is coverage from anchor positions chosen for external ranging. Run the footwell test before committing. If the answer is that one interior anchor has to be added, compare that added anchor honestly against a 60 GHz module rather than counting it as free.

In both cases, if a DMS camera exists, fuse it. Use it for seat occupancy classification, child seat orientation and false positive rejection, and let radar do the detection. That combination scores better and complains less than either alone.

And regardless of sensor, budget more for the warning cascade than you expect. The protocol’s scoring table shows systems with detection but without escalation scoring zero. More programmes lose points on the warning and intervention logic, which is software and vehicle integration, than on the sensor.

Where needCode fits

needCode builds in-cabin child presence detection on UWB radar, sensing a living presence by its breathing and micro-movement so that a sleeping or covered child is detected where a door sequence reminder is not, with AI classification to separate a child from a coat, a bag or a pet.

Given the recommendation above, the honest scope of that is narrower than a sales page would suggest. Where it is genuinely the right call is the vehicle already carrying UWB for a digital key, where a single sensing node can carry CPD alongside access, and where privacy constraints rule out an imager in a parked cabin. Building a child presence detection sensor this way draws on the same UWB engineering capability that produces the ranging stack, applied to the channel impulse response instead of the timestamp.

For teams still choosing, our free e-book on evaluating UWB radar for presence sensing compares UWB against cameras, PIR, 60 GHz radar and Wi-Fi sensing rather than arguing for one, and the compliance side of CPD is covered separately since the warning cascade is where most points are lost. When you reach validation, end to end testing covers running the protocol conditions as a repeatable bench rather than a one off demonstration. Talk to us if you want the footwell question answered against your own anchor geometry before you commit to a sensor.

Frequently asked questions