We compress the sequence of omni directional image based on IBR(Image Based Rendering), and render a large scale scene in real-time.
Compressed data is restored by product sum operation of eigen images and
the weight coefficient. This operation is all linear,
and because we run it in the fragment-shader of the graphics card,
real-time rendering is achieved.
Fig.1 : Omni directional image.
The omni directional images(ex Fig.1) contain redundant data
because the same object is captured from multiple view directions.
So, we consider efficient compression based on IBR, using images like Fig.2
Fig.2 : Image for compression.
We divide Fig.2 images into n*m block unit, and do tracking against each
block image. Then, KL expantion is done against the obtained block image set.
At this point, the high compressibility can be expected of the similar
block sets because of tracking by using EPI and Blockmaching.
Fig.3 : Compression method.
Fig.4 : Rendering algorithm.
Rendering is achived by restoration of all n*m block image.
The weight coefficient is the parameter corresponding to
the angle of the viewpoint and rendering board.
And, because the processing of Fig.4 left part(product sum operation)
is done in the frangment shader, high speed restoration and real-time rendering
are achieved. Fig.5 is the result of rendering.
Fig.5 : Result of rendring.
Publications
-
Ryo Sato , Shintaro Ono, Hiroshi Kawasaki, Katsushi Ikeuchi
"Photo-Realistic Driving Simulator using Eigen Texture and Real-Time Restoration Techniques by GPU"
International Journal of ITS Research, Vol.6, No.2, pp.87-95, 3.2008
-
Sato Ryo,Ono Shintaro,Kawasaki Hiroshi,Ikeuchi Katsushi
"Real-Time Image-Based Rendering System for Virtual City Based on Image Compression Technique and Eigen Texture Method"
IAPR International Conference on Pattern Recognition (ICPR08), pp.1-4, 12.2008
|
