Let's get serious about DMS. The footage of a Tesla slamming into an overturned truck on a highway in Taiwan has been featured in every automotive publication since early June...
It’s time to get serious about DMS — Driver Monitoring System.
The footage of a Tesla slamming into an overturned truck on a highway in Taiwan seems to have been featured in every automotive publication since it occurred in early June. Watching the video, three thoughts cross my mind:
One of the most extraordinary disconnects I have experienced in 20 years of covering automotive electronics is how many people entirely dismiss driver monitoring systems (DMS), even though Euro NCAP (New Car Assessment Program) and the updated European General Safety Regulations (GSR) both specify the need for active driver monitoring starting in just a couple of years.
Over and over when talking with supposedly smart and well-informed people, I am confronted with approximately the same viewpoint, which can be summarized as “DMS is at best an interim solution and at worst already obsolete.”
Rightly or wrongly, many people believe we are in the home stretch to L4 autonomy and that a “full self-driving” reality is but one over-the-air update away. Watch the Tesla crash footage again and honestly ask yourself this: How is that working out?
I don’t dispute that companies like Waymo and Cruise will offer some kind of autonomous shuttle or robotaxi service somewhere by 2025. But whether they can make any money from operating those services is a moot point, discussed in more depth in Walking on the Moon.
Also, I remain unconvinced that the general population is mentally prepared for any robocar to kill a child, which is a statistical inevitability as autonomous vehicles are deployed in greater numbers on public roads and highways. But these are all economic, legal and philosophical arguments and EETimes is a technical publication, so let’s leave those issues for others to debate elsewhere.
I recently saw this fascinating slide from Seeing Machines (Canberra, Australia), which is the company that provides the vision-based DMS for GM Super Cruise in the Cadillac CT6.
Let’s look further at what it tells us about some of the technical developments happening in the DMS world and the progress being made towards saving lives by making human drivers into safer drivers.
Blinded by the light
The most common criticism of the driver monitoring vision sensor in Super Cruise is that it can be blinded by sunlight and critics often use this to dismiss DMS technology altogether. I’m a total DMS geek, so I can tell from looking at the slide that the first two images on the top row use an 850nm infra-red (IR) optical path, while the other pictures all use IR components operating at a wavelength of 940nm.
How so? As the graph below shows, there’s substantially less energy in sunlight at 940nm than at 850nm, so changing the operating wavelength of the optical path to 940nm mostly eradicates the blinding issue. Thank you, Mother Nature.
Did I just hear you ask: “Why not start with 940nm components in the first place, then?” That’s because they weren’t automotive-qualified until the end of 2016, so on typical automotive timescales, they couldn’t have featured in production vehicles until 2019 at the earliest. Sure enough, 940nm vision-based DMS will start in series-production vehicles later this year.
If you are a tech company late to the DMS party and thinking of just throwing together a system for NCAP compliance, there is a further optical path challenge to be aware of: The quantum efficiency of CMOS image sensors is substantially lower at 940nm than at 850nm, meaning that much more powerful LEDs must be used to illuminate the driver’s face.
IR light might be mostly invisible to the human eye, but that doesn’t make it safe at high power — think laser eye surgery. This requires validation of the optical path as compliant with IEC 62471 for eye safety.
Your design choices as a DMS latecomer therefore are to use 850nm illumination and the sensor be blinded by sunlight, or to use 940nm illumination and potentially blind the driver. And you thought automotive-grade DMS was easy, right?
All of the design and development issues for vision-based driver monitoring operating at 940nm have now been resolved, but this is one reason why DMS has taken so long to validate for automotive use.
How to train your machine
Using face masks with a DMS is easy too, right? Well take a good look at the real-world examples in the associated video and you’ll see again why automotive DMS development is vastly more complicated than it appears. My favorite is the center bottom row, which looks like a facehugger from the Alien movies.
Humans can use knowledge-based reasoning to analyze these images and determine that they show human drivers wearing various types of mask. What this video perfectly demonstrates is that to successfully develop a DMS robust enough for use in real-life conditions requires massive amounts of naturalistic driving data to train the algorithms, combined with exhaustive testing and validation. This takes years of R&D.
In a Covid-19 world it is now not uncommon to see drivers wearing personal protection equipment, meaning DMS has to operate with complex combinations of full-face masks, safety goggles and respirators. Did you think of that?
State-of-the-art driver monitoring vision algorithms are now so extensively trained and validated that they require no calibration; track faces and eyes almost instantaneously; operate across a near 180-degree range (head turned fully to the left or right); and work through most sunglasses.
The algorithms are agnostic for gender, nationality and ethnicity, are trained to work with hats, baseball caps, jewelry, scarves, hygiene masks, safety glasses or goggles and also religious garments such as the hijab and niqab.
Everything self-driving in the long run
Anyone who believes DMS is an interim solution that will be replaced in the long run by self-driving vehicles is absolutely correct. But as the economist John Maynard Keynes wisely observed, “In the long run we are all dead.”
I’m an analyst, not a time traveler, so I don’t know the future of DMS versus self-driving and encourage you to make up your own mind. Just make absolutely sure you understand precisely what Euro NCAP and the European GSR directives say about the requirements for monitoring drivers for drowsiness and distraction.