Euro NCAP CPD is the assessment that decides whether a vehicle earns points for child presence detection, and since January 2025 it has stopped rewarding the systems most manufacturers actually shipped. Door logic reminders score nothing. Alerts that only tell the driver to check the back seat score nothing. The points now go to systems that detect the child, which means sensing a sleeping infant breathing at 30 breaths per minute, under a blanket, in a rear facing seat, in a locked car.
That is a sensing and classification problem with an automotive validation programme attached, and it lands on teams who often scoped it as a reminder feature. This article sets out exactly what the Euro NCAP CPD protocol tests, what the scoring change means in points, how the three candidate sensor technologies compare, and what the warning cascade requires of your vehicle architecture.
The figures below come from the Euro NCAP CPD test and assessment protocol itself rather than from vendor summaries, because the detail is where the engineering effort hides.
What Euro NCAP CPD actually tests
The protocol addresses heat stroke in parked vehicles. Its subject is a child up to six years old, defined for test purposes as up to 28 kg or 125 cm, and it covers four age bands, each with a specified child restraint system: a newborn of 4 kg in a rearward facing seat, a one year old in a rearward facing seat, a three year old in a forward facing seat, and a six year old on a booster cushion.
Coverage is all seat rows, including optional and removable rear seats. From January 2025 the assessment extended to the driver’s seat and to footwells. The luggage compartment is excluded.
The protocol draws a hard line between two ways of knowing a child is present. Direct sensing is the ability to detect the absolute presence of a human inside the vehicle by tracking heartbeat, respiration, movement, or another sign of life. Indirect sensing derives potential presence from logic: a rear door was opened before the journey, a seat pressure sensor closed, a capacitive pad registered contact. Every consequential requirement in Euro NCAP CPD follows from which side of that line your system sits on.
Three scenarios are assessed across nine use cases. Scenario 1 covers a child unintentionally left behind. Scenario 2 covers a child intentionally left behind, where data is collected but not scored. Scenario 3, assessed from 1 January 2025, covers a child aged three to six gaining unsupervised entry to an unlocked vehicle, which is the case that pulled footwells and the driver’s seat into scope.
The nine use cases vary the journey: different door opening and closing sequences, a child entering or leaving mid journey, a delay before the system activates to simulate a fuel stop, and checks that the system does not warn when the child has already left.
The scoring change is the whole story
Under the 2023 and 2024 Euro NCAP CPD scheme, an indirect system that produced an initial warning could earn a full point across all seats. That route is closed. From 1 January 2025, no points are awarded to vehicles with indirect sensing, or to vehicles equipped with an alert or an initial warning only.
Here is what remains available.
| Sensing and capability | 2023 to 2024, all seats | 2025 onward, all seats | 2025 onward, rear seats only |
|---|---|---|---|
| Indirect, alert only | not scored | 0 | 0 |
| Indirect, initial warning | 1.0 | 0 | 0 |
| Direct, initial warning only | 1.5 | 0 | 0 |
| Direct, initial plus escalation, Scenarios 1 and 2 | 3.0 | 2.0 | 1.0 |
| Direct, initial plus escalation plus intervention, Scenarios 1 and 2 | 4.0 | 3.0 | 1.5 |
| Direct, initial plus escalation, Scenarios 1, 2 and 3 | not applicable | 3.0 | 1.5 |
| Direct, initial plus escalation plus intervention, Scenarios 1, 2 and 3 | not applicable | 4.0 | 2.0 |
Two things follow for a programme manager. First, the gap between a direct sensing system with escalation and the same system with an intervention capability is a full point, and intervention is comparatively cheap: unlocking the doors, running climate control, pushing a notification through the connected car app, or routing to eCall Advanced or a third party service. Second, covering Scenario 3 is worth another full point, and Scenario 3 is where the sensing gets hard, because a child who climbs in unsupervised ends up in footwells and front seats rather than neatly in a restraint.
Under the 2026 rating structure Euro NCAP folds this assessment into its occupant monitoring group, and Smart Eye’s reading of the 2026 protocol is that CPD contributes up to five points there and that the direct sensing expectation is stated even more plainly. Confirm the current point values against the live protocol before you build a business case on them, since Euro NCAP revises these documents on its own cadence.
The detection problem is harder than the requirement sounds
“Detect a child” reads like one requirement. Euro NCAP CPD turns it into a set of physical conditions that a sensor and its classifier have to survive together.
The most demanding is a sleeping infant. The protocol specifies respiration rates by age: 30 breaths per minute for a newborn, 22 for a one year old, 20 for a three year old, and 18 for a six year old. A sleeping newborn’s chest displacement is on the order of a millimetre. Your sensor has to resolve that, at range, through the back of a rearward facing child seat.
Then the protocol adds occlusion on purpose. Testing uses a blanket of 70 cm by 90 cm in 300 GSM cotton polyester, and sun shades, to reproduce a child asleep and covered. Optical sensing degrades sharply here, which is a large part of why radar has become the default answer.
The child restraint systems are named rather than generic, so the seat geometry is not something you can design around: a Maxi Cosi Cabriofix for the newborn, a Maxi Cosi Pebble 360 with FamilyFix base, a Britax Roemer King II LS, and a Concord Vario XT-5. A rearward facing shell puts a curved plastic barrier between your sensor and the thing you are trying to measure.
Direct sensing systems also have to hold up across day and night lighting, across movement patterns from head and limb motion down to complete stillness, and they must not fire when they should not. False positive checking covers the case where the child left the vehicle before locking and the case where an object was placed in the rear. Compliance there is strongly recommended rather than scored, which is easy to deprioritise and expensive to retrofit, because a system that cries wolf gets switched off by owners and shows up in warranty data rather than in the rating.
If you are using an indirect system and it fails to warn when a rear door was opened before locking, the protocol requires a disclaimer in the owner’s handbook. That is a reasonable signal of how the assessment views indirect sensing now.
Choosing the sensor
Three technologies are in contention, and the choice is an architecture decision rather than a component decision, because it determines how many nodes go into the cabin and what else they can do.
| 60 GHz FMCW radar | UWB radar | Camera based occupant monitoring | |
|---|---|---|---|
| Vital sign sensitivity | Highest, resolves heart rate and respiration | Detects respiration micro movement, including under blankets and behind seatbacks | Cannot see a covered child |
| Cabin coverage | One sensor can cover the whole cabin | Typically needs more than one node for full coverage | Needs line of sight to each occupant |
| Footwell and occluded positions | Strong | Reported weakness, see note below | Weak |
| Shares silicon with other functions | Serves CPD, seat occupancy, intrusion and proximity | Same silicon does secure ranging and radar sensing by software configuration, so it shares the digital key platform | Shares with driver monitoring |
| Indicative sensor cost | Around 35 USD | Around 100 USD | Varies with the DMS programme |
| Maturity in production | Volvo shipped three 60 GHz sensors in the EX90 | Emerging, NXP and Qorvo both active | Mature for DMS, not sufficient alone for CPD |
The honest caveat belongs on the UWB row. Novelic, which sells 60 GHz radar, reports that UWB has difficulty detecting children in the footwell area and a high false positive rate, and states that UWB fails to perform well across all the Euro NCAP CPD scenarios needed for four points. Treat that as a competitor’s assessment rather than a neutral finding, and treat it as a test you should run yourself on your own cabin geometry before committing.
The case for UWB is architectural. Qorvo’s platform integrates RF, baseband and configurable processing so the same silicon supports secure ranging and radar sensing through software configuration rather than separate chips. If the vehicle already carries UWB anchors for a digital key programme, sensing becomes a firmware and algorithm exercise on hardware you have already paid for, certified and placed. The incremental bill of materials for child presence detection is then close to zero, and one node can also serve seat occupancy, seatbelt reminder and intrusion detection. Our e-book on evaluating UWB radar for presence sensing works through what that dual use actually requires.
The case for 60 GHz is performance and precedent. It resolves smaller movements, one sensor covers more cabin, the part is cheaper, and there is production evidence behind it. ABI Research expects vehicular child presence detection to drive 3.5 million 60 GHz automotive radar shipments by 2030, with driver and occupant monitoring systems growing at a 25 percent compound annual rate between 2023 and 2030. That is the shape of the in cabin occupancy detection radar sensor market as it stands, and it tells you where the tier one supply base is investing.
The decision rule is straightforward. If there is no UWB in the vehicle, 60 GHz is the lower risk path to points. If a digital key programme is already putting UWB in the cabin, run a feasibility study on sensing before you add a second radio, because the answer changes the sensor count and the harness.
Whichever radio you pick, the classifier is the part that decides whether you pass. Detecting a periodic micro movement is signal processing. Deciding whether that movement is a child, a pet, a phone vibrating on a seat, or a coat moving in a draught is machine learning on the edge, trained on data you have to go and collect.
The warning cascade is a vehicle architecture problem
Euro NCAP CPD specifies the alerting behaviour tightly enough that it constrains which ECUs stay awake and what the telematics unit has to do.
| Stage | Requirement | Timing |
|---|---|---|
| Occupancy assessment | System may take time to decide | Up to 10 seconds after locking |
| Initial warning, 2025 onward | Must be both visual and audible, from the vehicle, distinct from the normal locking signal | Minimum 3 seconds duration |
| Scenario 3 initial warning | Triggered by unsupervised entry to an unlocked vehicle | Within 10 minutes of door closure |
| Driver delay | Driver may postpone the warning, system must still fire | Up to 10 minutes |
| Escalation, direct sensing only | Exterior signal for 15 seconds or more, haptic to key or phone, app notification, or alert to registered devices | Starts no more than 90 seconds after the initial warning ends, repeats at least every minute for no fewer than 20 minutes |
| Re-trigger after cancellation | If the child is still detected after the warning is cancelled | Every 90 seconds |
| Intervention, optional | Unlock, reduce temperature, notify via app, eCall Advanced, or third party service | Within 10 minutes of locking, or 5 minutes after the first escalation |
Read that as a power budget rather than a specification. A system that must sense continuously and escalate every minute for twenty minutes, on a parked and locked vehicle, is drawing from the twelve volt battery while the car is asleep. Escalation is required only for direct sensing systems, and direct sensing is now the only route to points, so every manufacturer chasing this assessment inherits the same quiescent current problem.
There is a second architectural consequence. Escalation that reaches a phone means the telematics path has to work from a parked car, and intervention through eCall Advanced or a third party service means a defined interface to a service provider. Those are integration items with lead times, not software tasks. Teams that discover them during validation lose a gate.
The protocol also permits a status notification at ignition off, and then explicitly notes that it is optional in order to avoid habituation. That is a useful piece of guidance about the intent of the whole assessment: the driver is not the sensor.
What this means for a programme
A Euro NCAP CPD programme splits into four tracks that do not run in sequence.
Sensing hardware and placement comes first, because it determines everything downstream. Node count, mounting position and the harness are frozen early in a vehicle programme, and a sensor that cannot see the footwell cannot be fixed in software later. Validate placement against Scenario 3 geometry, not just against a child seat in the second row.
Classification and data collection runs longest. You need labelled data covering four age bands, four named child restraint systems, blankets and sun shades, day and night, stillness and motion, plus the negative cases that drive false positives. Collecting that data on production cabin geometry is the long pole, and it cannot start until the sensor placement is settled, which is why the two tracks have to overlap rather than queue.
The warning and intervention chain crosses domains: body controller, instrument cluster, horn and lights, telematics, the connected car backend and the mobile app. It needs an owner early, because no single domain team owns it naturally and it is exactly the kind of cross domain feature that surfaces at the integration gate.
Homologation and evidence runs alongside. For direct sensing the protocol requires manufacturer supplied validation data for the test tools, and a direct comparison between output recorded with humans and with the test tools in a vehicle environment. That evidence package is a deliverable with its own schedule.
If you are targeting the four point configuration, the sequence that works is: fix sensor placement against Scenario 3 first, start data collection immediately on that geometry, assign the warning chain to a named owner in the same week, and treat intervention as in scope from the start rather than as a stretch goal, because it is the cheapest point on the board.
Where needCode fits
needCode builds in cabin child presence detection on UWB radar in sensing mode, detecting the micro movement of respiration to find a sleeping or still child, with edge machine learning to tell a child from a coat, a pet or an empty seat. One UWB node can carry child presence, seat occupancy, seatbelt reminder and intrusion detection, which is the argument for reusing the digital key radio rather than adding a second one.
The team works on Qorvo QM33 and QM35 in sensing capable configurations as a certified Qorvo partner of more than eight years, plus NXP, STMicroelectronics, Infineon and legacy DW3000 designs. needCode is a Nordic Semiconductor Design Partner for EMEA and a UWB Alliance member, and holds ISO/IEC 27001:2022 and ISO 9001:2015. Our automotive and SDV practice covers digital key, secure ranging, in cabin sensing and OTA as one programme.
To be straightforward about evidence: needCode does not publish named child presence programmes, so what we can show is UWB and edge AI work in adjacent domains rather than a badged CPD deployment. Engagements usually start with a two to four week feasibility study on your cabin geometry, scored against the Euro NCAP CPD scenarios, which is the right size of commitment for a decision this architectural.
If you are deciding between adding a 60 GHz sensor and reusing UWB you already have, book a discovery call and we will look at your node placement.
Frequently asked questions
No. From 1 January 2025 the protocol awards no points to vehicles with indirect sensing, and none to vehicles equipped with an alert or an initial warning only. Indirect sensing means deriving potential presence from logic such as a rear door opening before a journey, seat pressure, or capacitive contact, rather than detecting a sign of life. Systems of this kind were worth up to one point under the 2023 and 2024 scheme, and that route is now closed. There is a further consequence for manufacturers who keep an indirect system: where it fails to warn in the case of a rear door opened before locking, the protocol requires a disclaimer in the owner’s handbook. If your vehicle is being assessed from 2025 onward and you want any points at all in this area, you need direct sensing plus an escalation cascade.
Direct sensing with both an initial warning and an escalation cascade, covering Scenarios 1 and 2, which is worth two points across all seats or one point for rear seats only. The initial warning must be both visual and audible from the vehicle, must be distinct from the normal locking signal, and must last at least three seconds. Escalation must begin no more than 90 seconds after the initial warning ends and repeat at least once a minute for no fewer than 20 minutes, through an exterior signal, haptic feedback to the key or phone, an app notification, or an alert to registered devices. Adding an intervention capability takes it to three points, and covering Scenario 3 as well takes it to four. Intervention is the cheapest of those increments, since unlocking the doors or running climate control uses systems the vehicle already has.
Not for the scenarios that carry the points. The protocol deliberately tests with a blanket of 70 cm by 90 cm in 300 GSM cotton polyester and with sun shades, reproducing a child asleep and covered, and it tests newborns and one year olds in rearward facing restraints where the shell of the seat blocks the view. An optical sensor cannot see through either. Cameras remain necessary for driver monitoring and useful for occupant position and posture in the wider occupant monitoring assessment, so most programmes end up running a camera and a radar rather than choosing between them. For the child presence assessment specifically, the sensing has to be radar or another modality that detects respiration through occlusion.
60 GHz FMCW resolves smaller movements, so it has the edge on vital signs, one sensor can cover the whole cabin, and the part costs roughly 35 USD against roughly 100 USD for UWB. It also has production precedent, with Volvo shipping three 60 GHz sensors in the EX90. UWB’s advantage is architectural: the same silicon can run secure ranging and radar sensing through software configuration, so a vehicle that already carries UWB anchors for a digital key programme can add sensing without new hardware, and one node can also serve seat occupancy and intrusion detection. The caveat to weigh is that Novelic, a 60 GHz vendor, reports UWB struggling with footwell detection and false positives. That is a competitor’s assessment, so the sensible response is a feasibility study on your own cabin geometry rather than accepting either vendor’s framing.
Scenario 3, assessed from 1 January 2025, covers a child aged three to six who gains unsupervised access to an unlocked vehicle, and it is worth an additional point. It matters for hardware because it extended the assessed area to the driver’s seat and to footwells, where the earlier scenarios concentrated on children secured in rear restraints. A child who climbs in alone does not sit where a child seat would put them: they end up in the driver’s footwell, across the front seats, or crouched between rows. A sensor placed and validated for a rear facing child seat in the second row will not necessarily see any of that. Because node count and mounting position are frozen early in a vehicle programme, Scenario 3 geometry has to drive the placement decision from the start rather than being validated against it later.
