OpenGL Image processing
I have written some code that provides me with the raw data of video frames and I want to manipulate the pixel data of each frame.
I am using OpenGL (in Qt) and C++.
After manipulating the video frame I want to render it using OpenGL.
I have researched that I probably should use FBOs to render the data, but I don't quite know how it all works.
Do I have to copy the video frame into a texture and then use that texture in combination with the FBO? How would I set that up?
I have found some tutorials about how to use FBOs in 3D, but they don't really explain the mechanics about it. I only need to set up a 2D environment to display the video frames, don't I.
Can anyone point me to some tutorials that actually explain things in detail and use plain OpenGL?
If all you want to do is render raw image then use glDrawPixels. If you want to do any hardware-accelerated stuff (like scaling, postprocessing), load the image with glTexImage (glTexSubImage for subsequent updates), and draw fullscreen quad with this texture. FBOs are for offscreen rendering, eg. the opposite of what you want to do (you can use FBOs to render 3d scene into a texture which you can then use for subsequent processing or read it back to system memory).
It has to be in real time (can't use glDrawPixels). That's why I thought I should use FBOs to do all the processing offscreen and then when I am done I would display it. I might also have to use shaders to do the processing.
I will look into glTexImage; hope this will work.
I will look into glTexImage; hope this will work.
You might be mistaken here to some degree. Yes, FBO are used to do 2D image post processing. BUT the beauty of FBO is that you can first render 3D models using the FBO as render target for the projected 2D image and then you can access the projected 2D image of the 3D scene as texture. This texture can then be read to and modified on the CPU and reuploaded on the GPU for presenting on the screen (slow performance). Or you can apply post processing effects by using various vertex and pixel shaders while renderind this 2D image as fullscreen quad while applying the effects (good performance).
Anyway, if you already have a 2D image there is no 3D rendering involved. Hence you do not have a need for an FBO. Just create a texture and load your image into the texture. You can then render a fullscreen quad with this texture while applying your effects as shaders. Or you modify the image data before uploading it to the texture. But since we are talking about a video here I would suggest doing as much as you can on the graphics card. Read: Use Pixel shaders to apply your post processing.
The bottom line is: A FBO is only of good use if you have either a 3D scene you have to create your 2D image from, first. And / or if you have several steps of post processing which you can't combine but whereas each step relies on the input of the preceeding step as start image.
Anyway, if you already have a 2D image there is no 3D rendering involved. Hence you do not have a need for an FBO. Just create a texture and load your image into the texture. You can then render a fullscreen quad with this texture while applying your effects as shaders. Or you modify the image data before uploading it to the texture. But since we are talking about a video here I would suggest doing as much as you can on the graphics card. Read: Use Pixel shaders to apply your post processing.
The bottom line is: A FBO is only of good use if you have either a 3D scene you have to create your 2D image from, first. And / or if you have several steps of post processing which you can't combine but whereas each step relies on the input of the preceeding step as start image.
You want to use Pixel Buffer Objects which allows you to direcly map data to DMA channels in video memory.
See
http://www.songho.ca/opengl/gl_pbo.html#unpack
See
http://www.songho.ca/opengl/gl_pbo.html#unpack
Quote:Original post by samson77
I thought PBOs shouldn't be used anymore and FBOs are preferred?
Maybe you are thinking about p-buffer (the 2000 extension)
http://www.opengl.org/wiki/P-buffer
PBO=pixel buffer object
One day, we'll add a few words about PBO in the Wiki.
[Edited by - V-man on September 9, 2009 6:19:50 PM]
The video I will be playing back is either 720x576 (PAL) or 1280x720.
So they are both non-power of 2. How do I go about creating a texture? Do I just take the next size up to be power of 2? How would I put that texture then onto a quad?
So they are both non-power of 2. How do I go about creating a texture? Do I just take the next size up to be power of 2? How would I put that texture then onto a quad?
Quote:Original post by samson77
The video I will be playing back is either 720x576 (PAL) or 1280x720.
So they are both non-power of 2. How do I go about creating a texture? Do I just take the next size up to be power of 2? How would I put that texture then onto a quad?
OpenGL 2.0 includes the GL_ARB_texture_non_power_of_two extension, which removes the power of two restriction, so having hardware that has GL 2.0 support at least will do.
Even if implementation does not support 2n textures, you can break up the video frame image into smaller 2n patches and load them via glBindTexture() and glTexImage2D(). As for doing subsequent updates, use the glBindTexture() and glTexSubImage2D() functions, as snoutmate mentioned.
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement