Revision history [back]

First of all, a FPS, which is frames per second of 0.04 is terribly slow compared to 0.15. I think you want to achieve a FPS of about 30, which means 1/30 seconds processing for each frame. If you can get a FPS of 20 then it will already start to look real time.

I guess you want to create a large image panorama, in order to get a large view? Or could you else be more specific of about what you want to reach or which code you are using (if it is too long, create pseudocode or code snippers).

Basically getting this real time would be very hard. Possible you will need to use GPU functionality to speedup your processing pipeline. It also depends on the image size you use as input. So really, we need more explanation to specifically help you.

Try to answer this:

• How large are my images?
• What is my exact application? Can I provide some demo pictures?
• Which step in my processing pipeline takes longest (debug using timers)? Is it feature detection? Is it feature descripting? Or is it feature matching? Or does the actual deformation takes longer?