Rear Mounted Radar: does such a product exist?

[spunkyj]

Many cars are now outfitted with "smart" sensors that help them park, or warn of objects below eye level, etc. I was wondering if anyone knows of a rear mounted radar/sensor for a bike that would warn of a quickly approaching car on collision trajectory. I'm not a fan of mirrors (and feel I can't judge a car's side-to-side position very well with them) but think it would be great to have a sensor that would warn me of a car approaching directly behind (and on a trajectory to hit me), rather than one that will just pass me.

A quick web search turned up nothing. Does anyone know of such a product?

[CB HI]

How heavy a bike are you willing to ride and I doubt it would have sufficient range to save you.

No products out there right now.

[dougmc]

I do not know of any such product. However, I question how effective it could be for the use you've brought up -- and I'm guessing not very.

For a car, when the (front mounted) device can control the brakes, it can work very well. It knows how effective the brakes can be (probably set at around 0.6 gs?), and it knows the relative speed between the car in front and itself -- so it can easily calculate when a collision is imminent and prevent it right before it becomes impossible to prevent. If it does it at just the right time, there will be almost no false positives.

But rear mounted, on a bike, that would be different. It could warn you right when a collision is imminent and unpreventable (by the car slamming on it's brakes) but this would be like a second before the impact -- you wouldn't have time to swerve either. And if it gave you a longer period of time to react, it would be giving lots of false positives -- you'd be repeatedly sending yourself into the ditch for situations where the guy saw you and was just going to slow down in the last second or two.

And if such systems did become commonplace, I can see drivers (the ******** who would honk at you from just a few feet away or throw something at you) becoming aware of them and deliberately triggering them just for the "lulz". Of course, they wouldn't hit you because that would be their fault and the police might frown on it, but it's not a hit and run if they never touched you ...

I guess you could just have the system warn you when a car is coming up behind you, like when it's five seconds or so behind you -- it probably won't hit you, but maybe the warning will be nice. I would expect such a system to be going off all the time (making it more annoying than useful), but maybe with very little traffic it might be useful.

Personally, I'd just suggest a mirror (yes, I know, you're not a fan) -- you can get used to the movement in the mirror too, no technology needed.

[spunkyj]

I see your point about the difficulties in implementing such a system. I really do wish I could get used to a mirror; I've certainly tried. But I find it more distracting than anything, and can't seem to judge where vehicles are relative to me very well. And the fact that they are adjustable means that apparent car positions change every time the mirror is nudged. And on top of that I tend to get "double vision" through the mirror no matter how its adjusted. I think that people have different experiences with mirrors based on the biological equipment they start with...

[Jamesw2]

it would be easier to put a small chain ring on an A1 Abrams.
The avoidance system would need to be able to steer you out of the way of the approaching vehicle to be effective. If you were going 10 mph when it detected a collision course and the vehicle was doing 70 mph. (102 feet per second) the instrument would need ( for a 1 second avoidance alert) 300 feet.

A car only needs 100 feet, or less, to change course to put it on a collision course. Line of sight also plays a part if the vehicle , going off course, is behind a non collision course vehicle how could it tell what the second vehicle was doing. a curve in the road also a factor

lidar radar and a few hundred pounds of computer hardware

[ItsJustMe]

The parking sensors on cars are ultrasonic, and are not that directional really.

You really would need quite an accurate system to know if a car was going to collide with you or not. Just like with optics, resolution depends on wavelength and the size of the sensor. For ultrasonic, you'd need a very large sensor. For radar you might be able to do it with a phased array antenna maybe a foot across but you'd need some hefty processing power to be able to drive that and generate the resolution you need, and between the processing and the radar transmitter you'd probably need at least a couple hundred watts to drive it.

The best bet might be a LIDAR sensor. Again, you'd need non-trivial processing, but the sensor would be smaller, I'm guessing that with some very careful engineering and some custom chips you could get away with something on the order of a laptop computer and maybe 100 watts of power, and the sensor would be a pulsed infrared laser and a spinning mirror like you see on top of the Google self driving cars. In the end I think at best you'd be looking at something which would be difficult to make stand up to the shocks involved with being mounted on a bicycle, and which would be extremely cumbersome and I can't imagine would weigh less than maybe 30 or 40 pounds.

Collision detection in the real world is non-trivial. The most advanced projects around for self-driving cars do it, but they have a trunk full of computers and an internal combustion engine to provide power, and a very stable platform to mount the sensors on.

The best sensors and processing that you're ever going to get are sitting on top of your neck already.

[unterhausen]

If you watch how motorists pass each other on the freeway, it's amazing that there aren't more collisions because a large number of people change lanes at the last second. I'm sure they pass cyclists the same way, and that would make detecting imminent collisions very difficult

[sggoodri]

lidar radar and a few hundred pounds of computer hardware

The sensing and processing technology need not weigh more than a few ounces using laser radar, IR pattern triangulation (think XBox Kinect) or stereo vision and purpose-built ASICS with embedded processor cores. I work with very small digital video cameras that perform computer vision on an embedded processor the size of my fingernail. Power consumption would be less than or equal to that consumed by a bike light, mostly for those solutions involving an active emitter; the imaging and processing would require only a couple of Watts. For instance, any of the triangulation based solutions could be implemented by leveraging the block matching components already built into the motion vector estimators of commercial video compression SOCs used in low power digital video cameras and IP cameras.

I agree that the main problem is deciding what to do with such information with so little time. By the time one knows that a collision is imminent, there might not be time for more than deploying an airbag, which may or may not be practical - airbags involve technologies I know much less about.

[Giro]

A recent issue of Consumer Reports had an article on the electronic vehicle to vehicle wireless communication systems under development. These are going to do things such as warn you about a vehicle ahead of you around a blind turn, etc. They also mentioned bicycles, although the article did not mention any testing of wireless equipped bicycles at present.

As a bicyclist, I have mixed opinions on this. If motor vehicles are routinely equipped or retrofitted with these, drivers will increasingly rely on them and tend to be less vigilant about other vehicles like bikes. On the other hand, if bikes routinely had these, it could improve safety. However, given the number of cyclists at night who do not even use lights, I am not optimistic about bicyclists routinely carrying such an electronic devices.

Some details about this on DOT's site.

[dougmc]

I don't see where it needs to be particularly large.

Ultrasonic might be the easiest way. Bounce a pulse back, see how long it takes to come back -- that will give you the distance. Look at the doppler shift of the signal you receive, and that will tell you the relative speed. Polaroid cameras were doing ultrasonic sonar based autofocus, what, 30 years ago? and did it in a small package cheaply. I imagine modern stuff could add the Doppler shift measuring and dividing the relative speed by the distance to determine the "threat" pretty easily. Any of those small embedded computers could provide more than enough computing power for just a watt or two.

You could make sure the ultrasonic pulses were aimed back in a tight beam -- that would help stop catching cars in another lane unless they were far back.

It's possible that radar would be as easy to do as sonar nowadays. Pick something with a small wavelength and you can have a small but still highly directional antenna. (Again, you want it pointing back as tight as possible so it doesn't pick up cars in the adjacent lanes.)

But as small, cheap and effective as you could make the device ... there's still the problem of when to tell you where you still have time to do something about it but not get false positives -- and I think this problem is impossible to surmount, at least for a rear mounted device on a bike.

But adding these things to the front of a car sounds like almost a no brainer -- it should be relatively easy to make them completely reliable, able to detect cars, pedestrians and cyclists and to prevent a collision at the last second. That would probably scare some people, but it would be less scary than actually hitting them, and since it waits until the collision is just about imminent it wouldn't be second guessing the driver (unless the driver wanted to collide, of course.) And of course, with a car the size and power consumption become much smaller issues.

[genec]

You guys are working the problem wrong... put the detector on the automobile and make them avoid cyclists. This is already being done in some expensive luxury cars that have mm wave radar systems to help them avoid things like deer.

Don't complicate the bicycle, it is near perfect as it is. Not to mention that avoidance of objects in front of the vehicle is the responsibility of the driver of that vehicle (be it computer or human). This is even written into law.

[genec]

I don't see where it needs to be particularly large.

Ultrasonic might be the easiest way. Bounce a pulse back, see how long it takes to come back -- that will give you the distance. Look at the doppler shift of the signal you receive, and that will tell you the relative speed. Polaroid cameras were doing ultrasonic sonar based autofocus, what, 30 years ago? and did it in a small package cheaply. I imagine modern stuff could add the Doppler shift measuring and dividing the relative speed by the distance to determine the "threat" pretty easily. Any of those small embedded computers could provide more than enough computing power for just a watt or two.

You could make sure the ultrasonic pulses were aimed back in a tight beam -- that would help stop catching cars in another lane unless they were far back.

It's possible that radar would be as easy to do as sonar nowadays. Pick something with a small wavelength and you can have a small but still highly directional antenna. (Again, you want it pointing back as tight as possible so it doesn't pick up cars in the adjacent lanes.)

But as small, cheap and effective as you could make the device ... there's still the problem of when to tell you where you still have time to do something about it but not get false positives -- and I think this problem is impossible to surmount, at least for a rear mounted device on a bike.

But adding these things to the front of a car sounds like almost a no brainer -- it should be relatively easy to make them completely reliable, able to detect cars, pedestrians and cyclists and to prevent a collision at the last second. That would probably scare some people, but it would be less scary than actually hitting them, and since it waits until the collision is just about imminent it wouldn't be second guessing the driver (unless the driver wanted to collide, of course.) And of course, with a car the size and power consumption become much smaller issues.

Exactly!

[squirtdad]

The sensing and processing technology need not weigh more than a few ounces using laser radar, IR pattern triangulation (think XBox Kinect) or stereo vision and purpose-built ASICS with embedded processor cores. I work with very small digital video cameras that perform computer vision on an embedded processor the size of my fingernail. Power consumption would be less than or equal to that consumed by a bike light, mostly for those solutions involving an active emitter; the imaging and processing would require only a couple of Watts. For instance, any of the triangulation based solutions could be implemented by leveraging the block matching components already built into the motion vector estimators of commercial video compression SOCs used in low power digital video cameras and IP cameras.

I agree that the main problem is deciding what to do with such information with so little time. By the time one knows that a collision is imminent, there might not be time for more than deploying an airbag, which may or may not be practical - airbags involve technologies I know much less about.

Of course the tape out on such a specific ASICS will be in milions of dollars, so the market has to be big. Might be able to do better with FPGA (Xilinx of course is the way to go.........shameless marketing plug) but then unit cost is an issue

[genec]

Of course the tape out on such a specific ASICS will be in milions of dollars, so the market has to be big. Might be able to do better with FPGA (Xilinx of course is the way to go.........shameless marketing plug) but then unit cost is an issue

Totally aside to the thread conversation at hand...

I work in a related industry, and I chuckle whenever we use that term "tape out" for either an ASIC or PCB (PWB more accurately) as the expression is so old, and yet is still common jargon. Honestly, when was the last time you or anyone you know delivered either a computer tape or a hand taped design to anyone for fabrication?

I know since the mid 80's I have delivered nothing but zip files... even before the internet, we sent zip files to BBSs maintained by fab houses... Just an interesting anachronism.

[ItsJustMe]

I don't see where it needs to be particularly large.

(snip snip snip)

You simply can't aim ultrasonic that tightly. The ability to create a tight beam is dependent on the wavelength and the size of the emitter and detector. The only way to keep it that tight is to use very high frequencies, microwaves at least, possibly up into optical wavelengths.

The difference between a car that's 100 feet back and is going to miss you by 3 feet and one that's going to hit you is miniscule.

I think possibly the best way to do this might wind up being image processing. Lidar is probably more promising though.

Really the best way, I still think, is to figure out how to use a mirror.

[squirtdad]

Totally aside to the thread conversation at hand...

I work in a related industry, and I chuckle whenever we use that term "tape out" for either an ASIC or PCB (PWB more accurately) as the expression is so old, and yet is still common jargon. Honestly, when was the last time you or anyone you know delivered either a computer tape or a hand taped design to anyone for fabrication?

I know since the mid 80's I have delivered nothing but zip files... even before the internet, we sent zip files to BBSs maintained by fab houses... Just an interesting anachronism.

OT for sure....... it seems like early terms seem to stick. Go into photoshop and many terms are old darkroom terms (dodge, burn in) I'm an IT apps guy... so I haven't helped with tapeouts for chips , but we still call them that. Over the years put silver sticky tape marks on tape, Cut tapes from Ibm mainframes and even did code with punchcards for a burroughs main frame. Maybe that is why C&V is a hangout

[genec]

OT for sure....... it seems like early terms seem to stick. Go into photoshop and many terms are old darkroom terms (dodge, burn in) I'm an IT apps guy... so I haven't helped with tapeouts for chips , but we still call them that. Over the years put silver sticky tape marks on tape, Cut tapes from Ibm mainframes and even did code with punchcards for a burroughs main frame. Maybe that is why C&V is a hangout

C&V?

I tape out designs on about a 6 week or so basis... wireless products of some kind... been doing it since the 80's.

[Looigi]

I dunno. I don't use my rear view mirror or listen for traffic behind me with the expectation of being able to take action and evade a vehicle that's going to hit me. I use them to see when it's safe for me to move left to avoid obstacles, avoid the door zone, take the lane for a left turn, etc. There are cars coming up behind and moving over a bit to pass me all the time. It only takes a second or two for one of them not to move over enough and hit me instead, in which case there'd be insufficient time to do anything about it, IMO. A car is going to be aimed at you until it moves over and an ultrasonic device could only report that they're bearing down on a collision course. You'd have to take evasive action for every approaching car.

[sggoodri]

Of course the tape out on such a specific ASICS will be in milions of dollars, so the market has to be big. Might be able to do better with FPGA (Xilinx of course is the way to go.........shameless marketing plug) but then unit cost is an issue

There are a number of video processing SOCs on the market that have software configurable logic to make flexible use of the hardware-accelerated computations like block matching search. The actual hardware fab doesn't need to change, just embedded software, so the cost of getting involved with custom development starts in the $10,000 range. I'm working on a video project that will process 1080p video at about 1 Watt. FPGAs are certainly another (good) way to go, but the cheapest way for a hobbyist to go is probably with a DSP, which of course will consume more power, but probably less than 10 Watts for this type of application.

[spunkyj]

I dunno. I don't use my rear view mirror or listen for traffic behind me with the expectation of being able to take action and evade a vehicle that's going to hit me. I use them to see when it's safe for me to move left to avoid obstacles, avoid the door zone, take the lane for a left turn, etc. There are cars coming up behind and moving over a bit to pass me all the time. It only takes a second or two for one of them not to move over enough and hit me instead, in which case there'd be insufficient time to do anything about it, IMO. A car is going to be aimed at you until it moves over and an ultrasonic device could only report that they're bearing down on a collision course. You'd have to take evasive action for every approaching car.

This is true in traffic, but not when you're on a highway shoulder or dedicated bike lane where cars shouldn't ever be directly behind you. But maybe these riding scenarios are too limited for such a device to be useful.

[dougmc]

You simply can't aim ultrasonic that tightly. The ability to create a tight beam is dependent on the wavelength and the size of the emitter and detector. The only way to keep it that tight is to use very high frequencies, microwaves at least, possibly up into optical wavelengths.

But it's wavelength that matters here, not frequency. (And yes, I'm aware of the relationship between the two.) That said, a 40 kHz sound wave and a 30 GHz microwave have roughly the same wavelength, so I would expect a similarly sized emitter/receiver to be needed in both cases to get a similarly tight beam.

But you're right -- the beam would have to be quite tight to fit in 3' at 100'. If needed we could increase the microwave frequency, but going higher on the sound frequency would increase the attenuation by air (which may already be a problem at 40 kHz). Some sort of IR, optical or UV laser might work better -- that could easily be as tight as we wanted it to be. I imagine the electronics needed would be more complicated than with the ultrasonics, but of course this is 2012, not 1978 (thinking of the Polaroid cameras again) -- I imagine it's still going to be small, cheap, and efficient.

Putting this on a bicycle is not going to be very effective, especially pointed backwards -- it would either give you lots of false positives, or give you a half second of warning before a serious collision, neither of which is particularly useful. But put on a car, aimed forward, connected to the brakes it could be very useful.

Detecting cars would be easy -- they're big and metal, reflecting ultrasonics and microwaves easily, and their license plates reflect lasers very well. Detecting bicycles would be harder -- the metal might reflect some ultrasonics and microwaves, and shiny bits might reflect lasers adequately. Pedestrians might be the hardest to detect of all -- little metal anywhere, clothes might be all black and therefore not reflect lasers well.

Detecting stopped cars/bikes/pedestrians might be a problem, as the reflected signal might be indistinguishable from the reflected signal from the environment around -- the road, curbs, etc. Having the system be able to stop a car going 70 mph from hitting something stopped might not be practical -- by the time the car realizes that a collision is imminent, it may not have time to stop, and no way would we let a system turn the car (not without a much much more elaborate system.) But even so -- if the car does slam on it's brakes before collision, that would certainly be better than nothing.

I guess in practice there would be a lever on the steering wheel to disable the system while the lever is pushed -- that would be used by drivers who like to change lanes at the last second before a collision, to stop the system from second guessing them. I'd prefer that such an override not be there (and people just stop doing that), but the drivers would probably demand it.

Interesting problem -- one I think that can be at least partially resolved by a technical solution.

I imagine the car makers have already worked all this out, as such systems are already available as options. Hopefully they're as effective as I think they could be, and they'll soon become as ubiquitous as ABS brakes are today -- they could save a lot of lives. (And a lot of cyclists if the system is sensitive enough to detect bicycles. "Run over from behind" collisions may be a small percentage of the total, but they are disproportionately deadly compared to other collision types.)

[siberia37]

I think the fallacy in this approach is assuming you can get out of the way of a car that's going to hit you. You can't, not with a mirror, not with an early warning (because of too many false positives). It's an objective danger of being on the road at all (cars get hit from behind too). Stop trying to change objective dangers and focus on changing subjective dangers. Make yourself more visible, stay off busy roads etc..

[ItsJustMe]

The biggest problem with using a technological approach is that what does it do? It's not going to be able to warn you any faster than you can see for yourself, and when you get the warning, you've got to react. Even if you are waiting for the signal and immediately dive into the ditch when it goes off, it's probably 2 seconds at least. Given that the thing will probably not go off that often, you'll spend 3 or 4 seconds thinking "what the hell is that noise" before remembering. Then you'll probably shoulder check to see what's going on. Diving into the ditch isn't a guaranteed good move either.

[genec]

But it's wavelength that matters here, not frequency. (And yes, I'm aware of the relationship between the two.) That said, a 40 kHz sound wave and a 30 GHz microwave have roughly the same wavelength, so I would expect a similarly sized emitter/receiver to be needed in both cases to get a similarly tight beam.

But you're right -- the beam would have to be quite tight to fit in 3' at 100'. If needed we could increase the microwave frequency, but going higher on the sound frequency would increase the attenuation by air (which may already be a problem at 40 kHz). Some sort of IR, optical or UV laser might work better -- that could easily be as tight as we wanted it to be. I imagine the electronics needed would be more complicated than with the ultrasonics, but of course this is 2012, not 1978 (thinking of the Polaroid cameras again) -- I imagine it's still going to be small, cheap, and efficient.

Putting this on a bicycle is not going to be very effective, especially pointed backwards -- it would either give you lots of false positives, or give you a half second of warning before a serious collision, neither of which is particularly useful. But put on a car, aimed forward, connected to the brakes it could be very useful.

Detecting cars would be easy -- they're big and metal, reflecting ultrasonics and microwaves easily, and their license plates reflect lasers very well. Detecting bicycles would be harder -- the metal might reflect some ultrasonics and microwaves, and shiny bits might reflect lasers adequately. Pedestrians might be the hardest to detect of all -- little metal anywhere, clothes might be all black and therefore not reflect lasers well.

Detecting stopped cars/bikes/pedestrians might be a problem, as the reflected signal might be indistinguishable from the reflected signal from the environment around -- the road, curbs, etc. Having the system be able to stop a car going 70 mph from hitting something stopped might not be practical -- by the time the car realizes that a collision is imminent, it may not have time to stop, and no way would we let a system turn the car (not without a much much more elaborate system.) But even so -- if the car does slam on it's brakes before collision, that would certainly be better than nothing.

I guess in practice there would be a lever on the steering wheel to disable the system while the lever is pushed -- that would be used by drivers who like to change lanes at the last second before a collision, to stop the system from second guessing them. I'd prefer that such an override not be there (and people just stop doing that), but the drivers would probably demand it.

Interesting problem -- one I think that can be at least partially resolved by a technical solution.

I imagine the car makers have already worked all this out, as such systems are already available as options. Hopefully they're as effective as I think they could be, and they'll soon become as ubiquitous as ABS brakes are today -- they could save a lot of lives. (And a lot of cyclists if the system is sensitive enough to detect bicycles. "Run over from behind" collisions may be a small percentage of the total, but they are disproportionately deadly compared to other collision types.)

I'm not going to blow holes in your technology... lets just say you have several mistakes and leave it at that... what you do have right is this aspect: But put on a car, aimed forward, connected to the brakes it could be very useful.

The bottom line is that the vehicle (operator) approaching from behind has the burden to avoid collisions with those in front of them, and such a device can be built that will in fact detect humans (mm wave "radar" is an example or LIDAR) and as a carrier and powersupply, the auto is a perfect platform. Such systems already exist and are options on some expensive luxury cars.

Here is a paper on one such system. http://www.fujitsu-ten.com/business/...l/pdf/15-2.pdf

Getting these to be standard equipment on cars is the real issue... as with airbags, it will take time.

Of course the real irony here is that every driver has an obstacle avoidance system, but they often chose to put it on "snooze..." their very eyes.