| Author | Message | ||
     clyde Rating: N/A |
Hi All, As you all may know, the next stage is already upon us in 3D... Auto-stereoscopy a.k.a ThreeD without the use of glasses. Many companies have made headway in that dept. like Phillips, Stereographics, X3D, and Sharp. But this has thrown up another problem. It was bad enough at the begenning, with normal 2view (lefteye/righteye) 3d...meaning there isn't enough content available right now! Now imagine for these auto-stereo displays to work, you actually need 8 or 9 Physical (or virtual in case of computer generated) cameras to make the 3D content!! This type of imaging or video is called Multiview video, and well isnt the easiest to create. Im starting this discussion, as im heavily involved with creating content for these displays and would like to meet similar people and learn/share techniques. To start off here's what I know: 1) you get "plugins" for popular software like 3Dsmax, maya etc that create the needed extra virtual cameras much like a Line of cameras all looking at a scene and then rendering 9 views of the same scene. (you need these 9views or more so that when you see the combined image on the autostereo displays, it looks almost like holographic 3d that floats outside and inside the surface of the displays) using just 2 views wont help as you will have to be standing head-on to see the effect and would'nt be able to move yr head even an inch sideways. 2) There is software that has been used for a totally different purpose .ie. to create smooth slomotion video in normal video shooting. Such software as Retimer by Realviz, Dynapel's SmoothMotion etc.. what these software actually do is "generate" in-between frames of video between two consecutive frames of video to create a smoother "slowmotion" effect.. of course this is used for normal video shooting like sports videos etc... But by using these software its possible to actually supply a stereo frame (leftview/Rightview) and these software magically generate the inbetweens my morphing the two source images. The effect is good depending on the scene, and you can specify how many inbetweens you want, but is not 100% in every instance. It does work very well in most scenes tho! 3) Creating inbetween views my usings Mpeg's builtin motion estimation of objects.... I have read several pages on this subject, on how mpeg codecs can actually be used track changes in a scene... what would be really cool is if someone wrote a software using this feature of mpeg to extract this data and generate inbetween views in scenes with any motion at all! 4)9 physical cameras... yes, you could in theory stick 9 cameras on a slidebar and record video. the problem is of course setting whitebalance, focus etc for all cameras at the same time. Syncing wouldnt be so much of a problem, as you could just use a clapper board at the begenning of the shoot as in normal video shooting. (yes die-hard stereo sync buff who read this board will argue about 1/10000 etc etc timing delays, but im wondering if these people actually have such accurate vision to see these mistimings :-) 5) Depth mapping. or Disparity mapping. This is a technique in which you look at a scene or frame of video and then manually assign shades of grey from white to Black to different objects in the scene to give them depth (black) to pull them off screen (white) with in-between shades allowing for different screen depths. Now again there is rotoscopy software that you can use to trace outlines on objects one at a time or many at the sametime and then color them the shades of grey over the lenght of the scene. Finally there is combining software such as 3DCombine or the software tools with displays such as from Sharp and Stereographics that will convert your flat 2D video to multiview videos by using the greyscale "depthmap" frames you have created. Now as you can imagine this is BlOODY time consuming work!  So what we need is 1) Some automatic object tracking and coloring software (3D combine has this feature, but isnt at a very mature stage compared to expensive rotoscopy software like combustion, shake etc) 2) A software that will take a normal 2view stereo video and use the numerous Stereo Disparity extraction alogrithms available on the net to create a "depth map" (such as ZKstereo) One such software is BasRelief but is used mainly for Lenticular poster creation (Same multiview principle but for still images) ..This ends my rather loooong post!, But hope to get further ideas and know-how from people with programming know-how that read this board and if anyone knows of additional software for doing depth map creation via automatic rotoscopy/tracking of objects in a scene. Or software that will do stereo mprphing to create inbetween views. Regards, Clyde | ||
| clyde Rating: N/A |
Ps> I hope this post has helped explain the basics to people new to auto-stereoscopy! | ||
| M.H. Rating: N/A |
Clyde: We had tested ALL the methods you describe in praxis (at LightpseedDesing and GALI-3D) ... The key point: The autostereoscopic monitors degradate the image quality a lot itself by the used principles. If you in addtion degradate the input data quality by some non 100% correct method of 9 view image aquizition, the result is BAD. In adition it is useful to "oversample" the data. it means that e.g. for 1600x1200 res monitor, the best result could be obtained form 4800x3600 9 view source (and not from 1600x1200). I thing that only two quality accpetable methods are following: 1) Realy calulate the 9 view in CG enviroment ... 2) For real live shooting use at least 3 cameras, and use the image analysis to get the missing 6 views. |
