Cars learn to spot what is hiding

A child asleep under a blanket presents a challenge for a car equipped with an interior camera, since the child simply disappears from view. At the InCabin Europe conference in Barcelona this week, Deep-In-Sight and Murata showed a system that could still detect the child by picking up the tiny movements of breathing through the fabric.

It was one of the more striking demonstrations of detection technology at the joint InCabin and AutoSens Europe events. In effect, cars are being tasked with tracking not only their passengers, but also hazards that headlights and cameras may miss. The technologies ranged from working prototypes to components intended for production vehicles.

The cabin gets more senses

Deep-In-Sight combines an occupant-monitoring camera with Murata’s 60 GHz radar. When the child is visible, the camera identifies the occupant and maps the seat; when a blanket or front seat blocks its view, radar checks for breathing. The company also demonstrated how the sensors could identify a passenger dangerously close to an airbag, or verify movement inside a parked vehicle.

The camera-and-radar system remains a proof of concept, with the companies seeking vehicle-maker and supplier programmes to take it into production.

Brighter Signals is turning the seat itself into a more discerning sensor. A conventional seatbelt reminder usually knows whether someone occupies the seat. The company’s thin fabric, placed between the foam and upholstery, maps weight across the cushion.

Chief executive Andrew Klein said it can distinguish a small adult from a large child, information that helps determine how an airbag should deploy. Fabric sensors placed elsewhere in the cabin could also detect a head approaching the dashboard before contact. Klein said the company has about ten pilots under way with carmakers and major component suppliers.

“They can all see, very few robots today can touch,” he said, drawing a link to the tactile sensing his company is developing for robots. Brighter Signals says its fabric can provide this more detailed occupant classification at about half the cost of existing seat technologies.

Aptiv, meanwhile, is working to extract more information from a camera already inside the vehicle. Its Advanced Occupancy Classification system uses that camera to identify the position and size of passengers, potentially replacing multiple seat sensors and their wiring. Aptiv puts the possible reduction in the relevant parts cost at up to 40%.

STMicroelectronics introduced a compact infrared image sensor, the SafeSense VD56GA, designed to make such cameras easier to fit in mainstream cars. Its small package can be placed behind a display, while greater infrared sensitivity helps it capture occupants when visible light is poor. ST said it has already shipped more than 10-million automotive image sensors for driver and occupant monitoring.

Smart Eye introduced an AI agent that combines voice commands with the driver’s gaze and information from sensors outside the car. “This agent is different. It’s adapted to the car,” said chief executive Martin Krantz. “I can look at a car outside of the vehicle, and I ask the agent, ‘What’s that?’” His gaze identifies the subject.

The assistant could also recognise when a driver was drowsy or distracted and create a dashboard widget on request. “The car doesn’t have to be the same as it was when you bought it because you can make your own applications inside of the car,” Krantz said.

Seeing beyond the headlights

Calterah showed how the radio anchors used by some digital car keys could also sense nearby objects. Its Dubhe chip coordinates them to detect a low kerb or an available parking space while retaining the key’s normal function.

Calterah’s EMEA field application engineering and marketing manager Yaohui Liu described squeezing into a narrow bay as warning tones erupt. “Every direction, every sensor starts to warn you, and you still want to park there,” he said. “Why? They beep because the measurement is not accurate enough.”

Calterah’s demonstration measured an obstacle’s height to determine whether the car would clear it.

Calterah also introduced longer-range radar chips and demonstrated a broken-down vehicle in a tunnel. The stationary car offers no movement clue, while the walls reflect radio waves. Calterah said its system detected the vehicle beyond 130 metres while avoiding a false braking response. As Liu put it, “Nobody likes their car braking for nothing, right?”

The Barcelona event showcased how thermal cameras can reveal a pedestrian or animal beyond the headlights.

Korean manufacturer Edge Foundry introduced ThermoK VGA, a heat-detecting sensor scheduled for mass production in December. Thermal night vision has appeared in cars since 2000, yet Edge Foundry estimates only about 0.2% of new vehicles have it. With cost being a major factor, its manufacturing process addresses both price and production volume.

“What has not been solved is making it at a price the market can absorb and at a volume an OEM can rely on,” said Edge Foundry senior manager of global business development Hyunkyung Park. “That is not a sensor design problem. It is decided by how the thing is made.”

French manufacturer LYNRED unveiled its CAPT640 thermal sensor for automotive cameras, designed to detect pedestrians in both darkness and headlight glare. It said it had invested more than €100-million in its Grenoble facilities for automotive production.

KD demonstrated an automotive camera connected by optical fibre over 40 metres. Its deputy general manager, Juan Pardo, said demand was coming from truck, bus and mining operators whose cable runs stretch beyond copper’s practical reach, but passenger cars with many cameras present another opportunity.

“We think it makes more sense to start with cameras, develop the ecosystem, validate the technology,” he said.

Testing the things people do

rFpro tests driver-monitoring systems in a simulated car with computer-generated occupants. One test has the driver keep their head facing forward while their eyes glance away from the road. Engineers call this “lizard movement”: the monitoring camera must detect that the driver is looking away even though their head barely moves. The simulation also reproduces the metal frames inside seats and child seats, which reflect cabin radar.

“Internally, we call these the invisi-bits of the car that you can’t see, but are there, and they’re affecting the radar,” said rFpro technical director Matt Daley. His team has measured how infrared light reflects from different parts of a face, down to the contrast between a nose and a cheek, so virtual cameras receive realistic images.

Solectrix showed how carmakers can test a camera’s image before committing to the electronics that will process it. Its proFRAME hardware captures and replays video from vehicle cameras, while SXIVE software lets engineers adjust the image processing and see the results in real time. The combination can help pinpoint the cause of a poor image.

“You can use the same sensor but use a different ISP, and you can really analyse where the problem comes from,” said Solectrix research and development manager Dr Roman Tzschoppe. The updates shown in Barcelona include new controls for configuring SXIVE and managing recorded video.

Aptiv introduced a platform for testing virtual cabin scenarios before a physical prototype is available. Managing director Craig Turner said validation was becoming the bottleneck as car software changes faster.

“Can we enable every developer with every single change to run a full set of tests?” he asked. Aptiv says the platform can cut validation costs by up to 70%. “Virtual testing is an extension of physical testing.”

Anyverse announced a programme to develop datasets and simulation models for the virtual assessment that Euro NCAP plans to introduce in 2029. Its chief executive, Víctor González, said the work involves measuring how closely a simulated test matches a real one.

“The goal is to say we have measured the distance between synthetic data and real data, so we can have some guarantee that when we use synthetic data for the assessment protocol, that will give us results we can rely on.”

* Arthur Goldstuck is CEO of World Wide Worx, editor-in-chief of Gadget.co.za, and author of The Hitchhiker’s Guide to AI – The African Edge.

The post Cars learn to spot what is hiding appeared first on Gadget.