Picture this: A roadway 3 lanes wide, every vehicle has radar. What prevents inter-vehicle interference? Do all the adjacent vehicles have time-domain sync with each other to prevent registering the pulse reflection from the vehicle in the next lane?
The European funding project MOSARIM (MOre Safety for All by Radar Interference Mitigation) started in January 2010 with the main objectives to investigate possible automotive radar interference mechanisms by both simulation and real-world road-tests and assess possible countermeasure and mitigation techniques in general guidelines and recommendations.
Don't know how present systems work, maybe interference is not such an issue with current level of deploiment. However, this can be pretty much avoided if pulse width is short, duty cycle is sufficiently low and and pulse timing is random. There will be sufficient consistancy in tracking objects that interference can be rejected.
I think it would be far more useful to detect the color yellow rather than red. I strongly suspect that the number of drivers who are simply unaware of a red light and drive right through it is far less than the number who are perfectly well aware of the yellow light and then step on the gas to try to beat it before it turns red.
A yellow light sensor aimed up at the traffic light would work nicely with simple computations that factor in the vehicle's speed and braking distance, and then alert the driver as to whether he or she can make the light or not. If not, perhaps even apply the brakes for him.
While there will always be incentive to simplify these systems due to cost pressure (especially as ADAS becomes a mainstream, high-volume technology), I believe that the most effective systems will combine multiple sensor types -- such as vision plus radar. And I think that when it comes to saving lives, we want the most effective systems. Over 1 million people die annually worldwide from automobile accidents.
In fact, I do understand that this is not an either or question. And yet, in talking about this with several participants at the conference here, I realized that there are many different shades of radars and vision technologies.
Carmakers can choose to use vision sensors integrated with more smarts and intelligence while adopting a lighter version of radar system. Or, they can pay more for the heavier- duty radar system and add a much more straightforward image sensor (sans too much intelligence). There seems to be a growing options for carmakers.
Another factor in favor of using vision in these applications is that a vision system can be thought of as a "software-defined sensor", which can be adapted to mulitple purposes. For example, Mercedes is using a camera and embedded vision system to scan the road surface and adjust the car's suspension in real time for each bump in the road, resulting in a dramatic improvement in ride comfort. See http://bit.ly/LUvH42 for a review of this technology.
For those who want to learn how such systems are built, there are still a few seats available at the Embedded Vision Summit on October 2nd in the Boston area, where we'll have a full day of presentations and demos on embedded vision applications, algorithms, design techniques, and technology. See http://bit.ly/1d3xTrK for details.
@JeffBier, thanks for the URL! Thinking of a vision system as a "software-defined sensor" is an intriguing idea. As more intelligence and smarts is getting integrated into the vision system, that seems to be the trend...
I know this is jumping way off topic, but I can imagine that ideally the sensors your car would be using wouldn't necessarily be mounted ON the car. Once we could have the infrastructure updated it could be pulling from aerial cameras and other stationary devices that see the big picture. A pileup a mile in the distance could be registered and compensated for long before your bumper mounted radar could do anything.
I know it is just a fantasy, but just imagine if we could implement control systems like this one that controlls these quadcopters into our roadways!
I also agree with the idea that the emphasis should be , putting whatever technology ( Vision /Radar) into the road infrastructure rather than on every car. That will improve the reliability of the whole system because the infrastructure management can be done with much more efficiency than the individual cars. The problems of dirt/uneven painting of lanes can be circumvented by redundant sensors in the infrastructure.
And as @Caleb Craft is saying, the infrastructure can show a driver the "big picture"
I disagree, Prabhakar. An infrastructure system would require involving numerous local governments, state/province governments, and national governments all working tofether. I believe the bureaucratic monster would ensure that we end with only half of an infrastructure and the side roads where many accidents happen would not be covered. I believe the divide and conquer method of deploying onboard systems at least ensures that it will be fully functional for some and eventually fully functional for everyone. The monumental task approach is more likely to half way work for everyone and be fully functional for no one.
Indeed, as jaybus0 stated, an infrastructure system is a huge public works project involving multiple government entities. Even if funding could be procured -- a highly questionable proposition -- the high cost & long schedule would not bode well for improving driving safety any time soon. It is far more effective to deploy safety systems on the vehicles themselves.
Definitely, that's the current mainstream thinking in the U.S. Many in the automotive industry are leaned towards putting more intelligence into cars (i.e. ADAS). Especially, under the current political and financial climate the U.S. is in today, that's a good bet.
On the other hand, some countries in Europe, and Japan (up to the point) are driving toward more a balanced approach -- smart infrastructure and smart cars.
There is a huge geographical diversity in their approaches to the emerging field.
In my opinion the additional money a car owner will have to pay for such driver assistance features installed and to be maintained in a car will be much higher than the expenditure to install a common driver assistance infrastruture.
I don't agree Prabhakar...in the long run perhaps...but for now there is no entity, corporation or government that could shell out billions of dollars needed to build the infrastructure so it needs to be done one driver at the time...Kris
Not only NOT a fantasy, but could happen faster than we might expect. There is comercial incentive to make vehicles, and their passengers, including driver, more connected to the web. It's just one more step to integrate driver assistance. With real time vehicle information sent to some sort of cellular network, the network could be sending heads up info back. The vehicles would still have a high level of autonomous responsibilty but would not be so blind to a larger picture. Fixed sensors could be later integrated.
Vision technology has advanced enough to reliability detect lanes and to identify objects in front. Challenge is if the paint to separate lanes are consistently there.
I agree with the assessment. I do believe driver assistance shall be an hybrid of both vision and radar simply because measuring distance is not quite reliable using vision based system. Rather, radar is doing a good job.
With the ADAS, who's responsible to an accident in case it happens? ADAS can be giving out warning signal. The liability becomes obvious. The challenge is when ADAS is actually doing part of the job.
It's all RF, so really we're talking about advantages of different ranges of wavelengths. If you want to read road signs, you can't do that with radar. If you want to see through thick fog, the optical spectrum becomes less than ideal. If you need to difrerentiate between cars clustered close together, say waiting at an inrsection, again radar might not be so good. Each car will look like a blob, and blobs will tend to blend together.
Also, not to be ignored is radiation hazard. I would caution against the notion that radar will be very long range. The last thing we need is to be bombarded with focused radar beam radiation the instant we get behind the wheel. Imagine traveling along a busy interstate with all cars around training radar beams at you, at short range. Or at stop lights. This will need to be carefully worked out.
Yes, although those would be a new type of road sign. I was thinking more along the lines of the way the police can now scan regular old license plates. You're talking about a new form of V2I, which of course may be more optimal for electronic sensing.
Given that most signs (in the US anyway) are standardized (think STOP signs) it should be an easy task for vision to pick them out (even in poor visibilty conditions - snow covered stop sign) and alert the driver. Even with a low quality GPS system I was made aware of both my speed and the "posted speed limit" in almost real time. If GPS was improved and kept up to date on speeds and hazzards then GPS would be a very easy way to be more road aware. That said, having driven a rental car with back up vision assist and blind side (and back up) radar I loved it! I would be willing to pay more for my next car with those features, esp. given my new / young drivers in the family.
ADAS is really a very required feature but it will be adding too much amount in the basic value of the vehicle, it is rightly said that ADAS will being a mandatory thing as it had happened in case of ABS and Airbags, but in India ABS and Airbags are still optional features. Mainly the adding of more electronics will be adding more players in terms of technology solution provider, I hope Mitsubishi is the only automotive mfg that has its presence in electronics segment at well, what Mitsubishi will doing in term of developing ADAS?
The local automobile manufacturer have a lot of influence on the local laws. Eg emission rules are very everywhere. But i agree that extra technology or gadgets are going to increase the overall ownership cost of the vehicle.
You are right so my point is, if the car manufacturer itself designs ADAS it may tend to reduce the overall price of the car, that's why I was asking about Mitsubishi as that is the alone company who is in electronic system development along with car manufacturing.
@vasanth kumar d: Yes. I do know about PIXY. Great project with innumerable applications...
I know -- I pledged to th eKickstarter campaign -- I cannot wait for my Pixy to arrive -- that's one reason why I'm working on learning the Arduino so i can use my Pixy on an Arduino-controller robot...
Radar is more real-time with less computational processing
Vision is more reactive and require computational grunt.
Personally, i take both! That's this field of data sensing call data fusion, if you want to reduce uncertainties and improve measurement confidences. You want both. Along with GPS, inertial, wheel RPM measurements i.e dead reckoning. Otherwise, it's not complete to me.
We already have a lot of vision assistance in vehicles eg reverse, side parking. Radar or any other system which can detect the collision trajectory of other vehicle/human much in advance or take action will be greately helpful.
You will not put an infrastructure only network into place. That still leaves the person to do as they will behind the wheel. You presume to much that the driver is the center point of this design change. The US government has already stated that as a prelude to autonomous vehicles, the vehicles will have to communicate to the infrastructure and vehicle to vehicle. The 2014MY will bring to market the first multi sensor vehicles. That is optical and radar. This is a typical product development cycle where you allow the sensor technology to mature. Both networks will be required. Optical can not see the stop sign behind the overgrown vegetation or know a stop sign has been knocked down. This is where the vehicle is pulling GIS information from the Cloud (vehicle to infrastructure) and reading internal network data and acting on it. What ever the operator costs are will not enough to mitigate city congestion and other factors looming.
As other writers have observed, radar and optical imaging have different strengths and weaknesses. It reminds me of Google Maps: having the street map, the satellite view, and the traffic overlay provide insights and navigational information that isn't available from any single channel. The radar may be very useful at alerting drivers to the car coming in from the right while the optical system may be very useful in alerting the driver to a yellow light that is on the verge of turning red. Of course this must all be overlaid with common sense. Having a green light doesn't make it right to cross an intersection - you may be "dead right" if some idiot is running the red light on the intersecting street. The challenge, therefore, is to integrate the various information channels and provide a composite overlaid output which is informative (not overwhelming) to the driver in real time. We also must ensure that a sudden splash of mud on the sensors doesn't "blind" a driver who, however unwisely, is driving in the dark depending upon on the instruments rather than his own view of the road.
The challenge, therefore, is to integrate the various information channels and provide a composite overlaid output which is informative (not overwhelming) to the driver in real time.
That is going to be a real challenge. The more I think about ADAS, the more convinced I am that we will need a separate driver's ed course -- just so that we understand what to expect and what not to expect from all the so-called safety features ADAS offers.
If a separate driver's ed course is required, the technology will be a failure for 2 reasons. First, information retention from driver's education courses is minimal (and legacy drivers don't need to take it when they buy a new car) and secondly, these features will be car model specific for a long time.
That said, some of the new safety features do have a rocky road of introduction. I suspect we all remember the first time that we drove a car with anti-skid brakes and they engaged. I thought the brakes were failing and pumped them to compensate. Especially as a business travelers who is exposed to many different rental cars during the course of a year, I don't know the repertoire of safety features on a car until they are invoked.
I think this driver assistance is going to require some kind of "head-up display" on the windshield that from the driver's viewpoint is simply an enhanced view of what they're already seeing. At most, perhaps this display will work like the headlights high beam switch to flip between views (and as the rear-view mirror flips between the day and night views). We will need to be sure that the display doesn't become so fascinating that it becomes a distraction...
Blog Doing Math in FPGAs Tom Burke 23 comments For a recent project, I explored doing "real" (that is, non-integer) math on a Spartan 3 FPGA. FPGAs, by their nature, do integer math. That is, there's no floating-point ...