After a successful KickStarter tour featuring showings at The White House, TechCrunch Disrupt, DARPA Robotics Challenge and countless flights around the US, CyPhy Works LVL1 Kickstarter has come to an end. We were unable to make the even a delayed delivery schedule due to the maturity of the sourced video components, especially EIS (Electronic Image Stabilization). All money and associated credit card and Kickstarter fees were returned.
You need to be a member of diydrones to add comments!
Well, Mark, "stabilized" is very different than the functionality of a gimbal.
Although not privy to the gory details, IS usually involves the combination of processing power and the right math.This is not unlike video compression, which first started by comparing the pixels from one frame to another and then saving only the CHANGES as opposed to totally new frames which are only 100 pixels different than the preceding or following frame.
Sony has been doing this for many years with good results - that is, IS in their sports and other cameras.
Their fancy systems seem to use a mechanical (IMU) setup:
That's a lot of work...and probably why it's on their fancy cameras.
I think the cats meow when it comes to drone and similar IS is to use a larger sensor and use only part of the sensor for each view. If you picture a curved sensor or lens, this might replace a gimbal.
The Parrot BeBop is probably the best current example of this technique. If you scroll down this page you will see a graphic of how they frame part of the sensor to produce an image.
So, in summary, it seems there are many different approaches to IS - what works on a hand held camera which is relatively stable anyway would have to be very different from one to replace a gimbal.
As far as IQ, IMHO that's can be overcome eventually by sensor size, processing power, etc - at least to a degree. But I think the higher end of drone cameras have zoom and interchangeable lenses making the Parrot type of IS difficult. These will probably use the current gimbal setup for a number of years - because as real photogs say - it's all about the "glass" (lens).
I was wondering about how and when gimbals will be replaced for some time now and at least get to the point where we can have 1080p electronically stabilized. My knowledge is limited here but could somebody explain how these guys are doing it and what shortcuts in quality the camera will be making?
I note that the other Gary's UAS news sidebar has a article on still another multi (Flystar ProEagle) biting the dust on Kickstarter or at least being suspended a tiny bit shy of their million dollar goal.
Apparently according to Kickstarter posts due to violations of the Kickstarter reward guidelines.
Looking at what they have offered I am once again skeptical that is the whole story.
For them to deliver what they have offered at the price they have offered it would seem to me to require for them to operate more economically than DJI or any of the Chinese companies.
Personally I think they undercharged and are now facing getting overrun by the cheap Chinese multis.
But hey, maybe they will resolve their rewards offer, placate Kickstarter and carry on.
I think the smart money would be on not likely on that one though.
Certainly a sufficient video chip is one part of the problem, but the fact is every year now they seem to be nearly doubling both the sensitivity and pixel density of video chips.
Will the very top end video capabilities at any given time be realized by EIS - no almost certainly not, but we are already at 8K pro level and beyond with extreme wide dynamic range.
And most of that is complete overkill even for most professional uses.
Using EIS for total stabilization will considerably degrade the potential of any given chip.
However, that said, I am virtually certain that within a year - two at the most very high quality 1080 P will be realizable on a full EIS camera. It will almost certainly use a 4K or more likely larger chip to produce it.
There will always be uses for high end ultra pro capabilities which will benefit from being able to take full advantage of the video chip rather than having the bulk of it tied up with EIS decoding space.
And those will use optical or mechanical stabilization.
But the fact is that EIS will eventually take over at least 90 percent more likely 99 percent of all video camera uses.
Frankly we are hitting the point beyond 4K where additional pixels is of very limited application and the emphasis has already shifted to ultra wide dynamic range.
For lower but still very useful and marketable resolutions like 1080P EIS is much more economical than all those motors and mechanical parts. (as well as being much smaller and lighter.
The Sony linked below actually stabilizes quite well at 1080P (Note it is a 4K camera but only stabilizes in 1080P by using all the 4K pixels.)
It is probably the current state of the art.
From what I have seen not quite there yet, but really quite close.
But it's not a computing problem, it's a physics problem. More pixels on a chip will only decreases the light sensitivity (less surface area per pixel). And if you can't get blur free data under difficult conditions from a high end 35mm sensor in a professional camera, how can you expect a cheap 1/4" sensor to behave any better? "Crap in, crap out" is a mantra you will often hear being used by photo/video engineers. No amount of computing power can take blurry sensor raw data, 'stabilize' it and make a good result.
I respectfully disagree.
It is certainly true that it is not an insignificant challenge (a concept which I think has plagued early attempters).
But the reality is that small scale EIS is getting better very rapidly now and for UAS use the main obstacle is roll (circular) stabilization around an indeterminant and moving point.
You also need to use a chip with not only extra pixels around the edge, but at least 4 pixels (preferrably more) for every output pixel.
If the video chip can make it possible, the only other thing you need is a sufficiently fast and parallel multi GPU microprocessor to extract the image into reality.
Honestly, from the current examples (most notably the newest Sony Action cams) I'd say they are about half way there.
As for vibration, while of course that is also compensated for by the stabilization method, decent multicopters used for video should have very strong vibration damping anyway.
My honest guess is we will see the first truly successful 1080P electronic stabilized copters within a year.
4K at pro levels will definitely take longer.
The thing is.. you don't exactly have to be a rocket scientist to see that 'electric' image stabilization (EIS) as the main method for image stabilization is not going to work very well on a UAS.
EIS simply put works by analyzing the image sensor data and moving the image around to remove unwanted motion. Obliviously the resulting image quality is then directly tied to the quality of the captured sensor data. And what are sure things to cause trouble for image sensors? Fast motion and vibrations. Both causing motion blur and rolling shutter in the exposure. And what are you guaranteed to find on a UAS? Fast motion and vibrations..
But even that is achievable if the stakes are high enough.
I have no doubt that eventually the electronics will get good enough for stabilization. But there's still no replacement for good glass.
I don't know if gimbals will become obsolete, but for most uses they will I am sure be replaced by electronically stabilized cameras.
Although the roll compensation is really intense, it is really just a matter of developing specific multi-GPU processor built just to handle the stabilization decoding that is fast enough and combining it with a video chip with sufficient extra pixels to let the reconstruction appear seamless.
Ultimately electronic intensive hardware is always smaller and cheaper and more appropriate for consumer use than involved mechanical architectures.
The electronic solution will always ultimately and economically win.
Let alone our multicopters, this is needed for regular cameras and cell phones, not a question of if, only when.
At least, the Oculus stayed true to its KS backers and is still a very public and open project.
I love my DK2 and will be getting the new one soon now.