You can run but you can't hide...
A team of engineers has developed algorithms that reconstruct images of objects hiding around corners or behind walls, and they believe it could be used to help make self-driving cars safer one day - albeit very slow ones at present. The researchers call this “non-line-of-sight imaging” (NLOS) and have described the technique …
Car and Driver magazine figured out a simpler way back in the 1980's. If you have a strong enough infrared lamp pointing straight ahead on the front of your car, it will swamp the police laser radar unit. The downside was that you had this huge, ugly, unaerodynamic lamp apparatus on the front of your car so the cops would know something was up.
I remember that article!
I could see it working for trucks, which often carry large off-road lights anyway. With the right filter the light would appear to be off. Probably easier to just not speed, as cool as the idea of a "stealth" car is, especially since there are lots of other ways to catch speeders.
Its hard to see how a system that depends on precision optics is going to work for very long in the real world of vibrations, temperature variations, and impacts from various insects.
I note that the researchers want to reduce the processing time down to seconds. You can travel quite a distance in a few seconds - they need to get it down into the millisecond range for it to be of any use for autonomous vehicles.
"yes, the real world is so different to a lab. Speed of processing data I think is always going to be key in the real world"
I suspect signal to noise ratio will be the key in the real world. Very few photons will be scattered around corners and then fewer still are scattered back to the detector. In the real world especially in daylight but also with other light sources providing noise I am sceptical enough photons can be gathered in a sensible time period to be able to reconstruct anything useful. OK you can take the difference between with the laser on and off but noise in the signal and noise in the environment will still add. Few photons mean noisy data. Reconstruction will if anything amplify the noise. This a much more fundamental limit than processing time.
Infrared will fry eyeballs too. Just because we can't see it doesn't mean it can't heat up our retinas. In fact infrared is worse, because unlike visible light it doesn't trigger our reflex to look away.
Current LIDAR systems are already only OK because of their constant scanning -- if the beam ever stopped on someone's eye for any length of time, there could be trouble.
That of course is why laser printers don't work (except for the insect impacts).
By the way, the response time of the human brain to an optical input is of the order of 200ms. (It's about 90ms for sound). When you say "millisecond range" I hope that's the timescale you are thinking of.
You know, slow down a little when going around corners that might be hiding something bad like a rockfall in the middle of the road, a deer ready to jump out, or whatever to leave sufficient room to stop. Like us meatbags
do are supposed to do.
But no, let's create some nearly impossible to perfect technology to allow us to go around curves 10 mph faster...
Speed around curves has more important limits. First physics, not of light but of centrifugal force and friction. Secondly, and more importantly, passenger comfort - reasonably priced sports cars on dry roads with new tires are able to slightly exceed one g of lateral acceleration, but the commonly accepted value for passenger comfort is only .25 to .30 g.
People might like to exceed one g when driving themselves for fun - when they are in control. But even then generally don't when they are on a point A to point B trip. Hardly anyone is going to be happy pulling hard turns in a car they aren't in control of.
...or, perhaps, just note that there is some sort of junction ahead you cannot see around and behave as if there was something there on a collision course with you every single time; slow down to a speed that lets you stop in a shorter distance than you can see ahead. It's what I do. And more often that I feel comfortable with there actually IS something approaching that may or may not have caused trouble otherwise...
I learned this lesson as a teenager, after a scare in which I barreled around a blind curve at 50 mph to find there was a tractor there doing 5.
I try to drive such that I can always stop within the distance I can see ahead. This actually lends itself pretty well to "slow in-fast out" cornering, since as I go through a curve my view opens up ahead.
Not the song.
I've walked along twisty, narrow country roads with no pavement (sidewalk in Merkin). More than once I've had to dive into the hedge as some car came speeding around a blind corner. These were roads that often have people on horseback riding along, and if the car had encountered one of those it would have ended nastily.
This see-around-corners trick, even if it worked in real time, would be no good there. You're not going to get any useful data bouncing light off a hedge. Or at a road junction where one corner has buildings (obscuring your view of the turning you wish to take) but the other corners have nothing (not even a hedge or fence).
If you could make it a lot faster, and a lot more reliable, it might be useful as a belt-and-braces solution. A second opinion that occasionally comes into play and improves safety a little in some rare situations. It would be better, at this stage, putting the money into solutions that work in all situations (otherwise known as driving safely).
Except not smooth with a reflective background but rough with a potentially non-reflective background.
I can see this working in USA cities with grid pattern roads and glazed shop fronts because meat sacks can do a very similar trick. Using a rough barked tree as a mirror would require serious illumination coupled with a very high ability to discriminate between reflections from the target and from noise.
I assume that you can tag the light (or other wavelengths) you emit so you can detect which incoming light is a reflection.
I would have thought the first stage would be to detect the reflections round the corner from other light emmiters. Such as the meatsack trick of seeing car headlights reflected in the store front on the corner. Can cars do even that yet?
The biggest issue I have with the research that's being done on autonomous vehicles, is that there is entirely too much time being wasted by these eggheads, on utter nonsense like this!
Perhaps in 100 years or so, after we've got quantum computing perfectly nailed down, research like this will make sense--but right now, it's not doing anything to help get these things out on the roads any faster.
We're already losing a race against the clock folks. The gray menace is already out there on the roads--clogging up traffic, and plowing into everything in sight, and it's only going to get worse as these baby-boomers keep getting older! After 70, fatal crashes start ramping up, and by the time a driver is in their 80's--they're as bad, or worse than teenagers!
[Technically, teenagers have a higher percentage of overall crashes--but tend to be involved in far more single-vehicle crashes (endangering only themselves or occupants.) Seniors on the other hand, are only slightly lower overall percentages, and tend to be involved in more multi-vehicle crashes, and multi-vehicle intersection crashes. (endangering everyone else around them.)]
We don't need these autonomous vehicles to be able to see shit that's hidden behind other objects, we just need them to to see better than an octogenarian with cataracts, we need them able to avoid imminent collisions faster than someone crippled by arthritis, and interpret signs/traffic signals better than someone with early onset Dementia/Alzheimer's!
These things are just trains without tracks, and once people start understanding that--they'll start putting time and money into figuring out the infrastructure changes needed to accommodate them. (i.e. pedestrian bridges over busy intersections, gates like the one's used at railroad crossings, cameras and/or sensors that will help to detect things beyond the car's field of vision.)
As it stands, I think we're already there, and the sooner people wise up and realize that the reason any of these things have been involved in collisions, is because of stupid humans doing unpredictable shit. Which means the sooner that human drivers are no longer the majority, the safer we'll all be. In fact, I'd be willing to be that autonomous vehicles (or just cars, as we'll be calling them soon) won't even need to be as complex or sophisticated with their imaging systems--because all autonomous vehicles will be abiding by the same rules, without exception, and once they can call talk to each other, they'll be able to provide each other with real-time updates about traffic conditions, or hazards on the road, like fallen trees or humans operating vehicles manually.
"he gray menace is already out there on the roads--clogging up traffic, and plowing into everything in sight,
I'm sure that you have bunch more stereotypes to throw into the mix.
"i.e. pedestrian bridges over busy intersections, gates like the one's used at railroad crossings, cameras and/or sensors that will help to detect things beyond the car's field of vision."
And are you volunteering to pay for all this infrastructure you seem to think is required?
Perhaps if the cops and those, supposedly, studying all this addressed the actual driving techniques and attitudes, rather than taking the easy route of speed cameras, they just might have a better result.
Here in New Zealand, the "authorities" have decided that a tolerance of 4 KPH is all that they will allow. No context required. A quick squirt of throttle to overtake or merge into another lane before slowing down again? Tough, you're pinged. Yet they keep banging on about driving safely. I get the feeling they have no idea what driving safely is. Just keeping below an arbitrary speed limit is NOT necessarily safe driving.
Biting the hand that feeds IT © 1998–2020