SoylentNews is people
SoylentNews
Arthur T Knackerbracket has found the following story:
The number of cameras in cars is increasing. However, through the flood of data the internal networks are being pushed to their limits. Special compression methods reduce the amount of video data, but exhibit a high degree of latency for coding. Fraunhofer researchers have adapted video compression in such a way that a latency is almost no longer perceivable. It is therefore of interest for use in road traffic or for autonomous driving. This technology will be on display at the Embedded World from 27 February until 1 March 2018 in Nuremberg in hall 4 (booth 4-470).
[...] The Fraunhofer HHI, for example, has made a decisive contribution to the development of the two video coding standards H.264/Advanced Video Coding (AVC) and H.265/MPEG High Efficiency Video Coding (HEVC). "With these methods, the data quantities can be sharply reduced. In this way, more than ten times the quantity of data can be transmitted," emphasizes the group leader of the "Video Coding and Machine Learning" department at the Fraunhofer HHI.
Normally, 30 to 60 images per second are sent from a camera to the vehicle's central computer. By compressing the image data, a small delay in transmission occurs, known as the latency. "Usually, this is five to six images per second," explains Stabernack. The reason for this is that the methods compare an image with those that have already been transmitted in order to determine the difference between the current image and its predecessors. The networks then only send the changes from image to image. This determination takes a certain amount of time.
"However, this loss of time can be of decisive importance in road traffic," says Stabernack. In order to avoid latency, the professor and his team only use special mechanisms of the H.264-coding method, whereby determining the differences in individual images no longer takes place between images, but within an image. This makes it a lowlatency method.
"With our method the delay is now less than one image per second, almost real time. We can therefore now also use the H.264 method for cameras in vehicles," is how Stabernack describes the additional value. The technology was implemented in the form of a special chip. In the camera it compresses the image data, and in the on-board computer it decodes them again.
(Score: 3, Interesting) by Grishnakh on Friday February 23 2018, @09:44PM (9 children)
I looked these up and VC-2 seems to be known as Dirac, and has high compression using wavelet technology, but doesn't appear to be designed for low latency, which is what's really important in this application, not extremely high compression. The motion JPEG2000 standard seems to aim at being an archival format, not high-performance. TICO, however, seems to fit the bill, but it doesn't even have its own Wikipedia page so it doesn't seem to be highly deployed. This "story" is really a press release from Fraunhofer, which is the same "institute" that came up with MP3 and other patented technologies, and makes money selling licenses. So they're basically coming up with a competitor to TICO it seems, and simply ignoring the competition.
(Score: 2) by bob_super on Friday February 23 2018, @10:47PM (8 children)
You are likely correct about Frauhofer press release.
As far as the three codecs I mentioned, read that sentence again: TV productions facilities are buying those, on my platform.
We got our JPEG2000 end-to-end, SDI in to SDI out to under 4 frames of latency (some trade-offs required).
TICO and VC2 let you run 1/4th compression at low latency, bring 4K to the bandwidth or raw HDp50 (~3Gb/s). We used that in remote production demos.
(Score: 2) by Grishnakh on Friday February 23 2018, @11:01PM (7 children)
We got our JPEG2000 end-to-end, SDI in to SDI out to under 4 frames of latency (some trade-offs required).
That seems slow to me. At 30Hz, 4 frames is 133ms. At 65mph, that's almost 13 feet, and that's just for the compression and decompression.
(Score: 2) by fyngyrz on Friday February 23 2018, @11:13PM
And of course there are full-analog HD cameras.
You want to examine a frame, capture it (takes exactly one frame time) and have at it.
My entire security system is analog 1080p; 16 cameras, no network load at all. What a vehicle needs on top of that is good capture hardware. Not a problem, technically speaking.
(Score: 2) by bob_super on Friday February 23 2018, @11:21PM (5 children)
Well, you can set JPEG2000 for ultra-low-latency, where it compresses pieces of the image as they are acquired, rather than compress the whole image (which implies 2 frames of latency right there). That would help you.
On the other hand, if you're driving 65MPH and 133ms/13ft imager-to-processing latency is a major issue, maybe, you shouldn't be driving 65MPH right there. Multi-second human reaction-action times are a lot longer, so that should leave a decent amount of time for computer processing/decision.
(Score: 3, Informative) by deimtee on Saturday February 24 2018, @02:45AM (4 children)
If you have multi-second reaction times you should not be driving at all. You probably shouldn't be allowed on the sidewalk unsupervised either.
Fast human reactions are on the order of a tenth of a second. Slow people may take as long as half a second, and IMO they are dangerous drivers.
If you cough while drinking cheap red wine it really cleans out your sinuses.
(Score: 2) by bob_super on Saturday February 24 2018, @02:56AM (3 children)
What's so hard about reading? I wrote "reaction-action".
Idiot jumps in your lane, you need to see the problem, recognize the problem, choose a solution, and move your hands and feet to execute the chosen response.
That takes seconds (1, 2, 4... how focused were you?), yet people drive safely every day at pretty high speeds, so multi-seconds is usually okay.
The 100ms acquisition of video has to be followed by the same steps in the multi-GHz silicon brain, which then has ultra-fast link to actuators. Is that latency the critical path that breaks the system? If compression allows a 4K resolution acquisition instead of HD, is the extra information worth the extra latency?
(Score: 3, Informative) by deimtee on Saturday February 24 2018, @06:57AM (2 children)
I stand by my figures above. A tenth of a second is a fast reaction, on the order of a professional racedriver. Half a second is slow, not uncommon, but a poor driver. If you have multi-second reaction times you should not be driving.
Or are you using the unusual term 'reaction-action' to indicate the total time from stimulus to completion of the action you take as a reaction to the initial action? For some reactions such as sustained braking that might take you into multi-second times.
If you cough while drinking cheap red wine it really cleans out your sinuses.
(Score: 2) by darkfeline on Tuesday February 27 2018, @04:54AM (1 child)
First, not everyone is a professional racedriver. That right there already brings you up to 0.2s.
Second, no one is capable of sustaining their optimum reaction time throughout an entire driving session, every single time they drive. Now you're probably up to 0.5s as a safe average
Third, while you may be able to "react" and press a button in under 0.2s in ideal conditions, the common actions when driving take far more time. A human cannot turn the steering wheel a good 180+ for a swerve in 0.1s, or move his foot from the gas pedal to the brake and depress it in 0.1s, even assuming 0s reaction time. Now we're up to a solid 0.7-0.8s in unrealistically ideal conditions.
Fourth, you seem to be under the false impression that most people aren't poor drivers. I hate to break this news to you, most people are terrible drivers and they should not be driving.
Fifth, you seem to be under the false impression that fast reaction times is what makes a good driver. As Frank Borman said, "A superior pilot uses his superior judgment to avoid situations which require the use of his superior skill." A good driver avoids situations that would require your idealistic 0.1s reaction time. If you're getting into situations that require your blessed 0.1s reaction times, you are a very bad driver.
Join the SDF Public Access UNIX System today!
(Score: 2) by deimtee on Wednesday February 28 2018, @08:24PM
The comment that started this said "multi-second" and was comparing human to computer reactions.
First I want you to get a stopwatch and time exactly how long two seconds is. At highway speeds you have travelled about sixty or seventy metres.
1. I agree that not everyone is a prodessional driver. That's why I said reactions vary up to half a second.
2. Again, I don't see the argument here. An average of half a second is still not multisecond.
3. If you are going to include the time the physical action takes then we are also comparing the efficiency of the controls, not just reaction times.
4. Most people are not as good as they think, but I don't think they are quite as bad as you seem to think. They would all be wrecked and off the road if they were.
5. Fast reactions are necessary but not sufficient. Not necessarily fighter pilot reactions, but if you consistently take more than half a second to realise that the car in front of you has started braking you are going to be involved in a lot of rear-enders. I mostly agree with the quote but sometimes the skill will be necessary, otherwise you would just have your best chess players flying your fighters.
If you cough while drinking cheap red wine it really cleans out your sinuses.