Getting better performance using OpenCV?

Question

I need real time processing, but the internal functions of OpenCV are not providing this. I am doing hand gesture recognition, and it works almost perfectly, except for the fact that the resulting output is VERY laggy and slow. I know that this isn't because of my algorithm but the processing times of OpenCV. Is there anything I can do to speed it up?

Ps: I don't want to use the IPP libraries so please don't suggest that. I need increased performance from OpenCV itself

Answer 1

Traditional techniques for improving image analysis:

1. Reduce the image to a monochrome sample.
2. Reduce the range of samples, e.g. from 8-bit monochrome to 4-bit monochrome.
3. Reduce the size of the image, e.g. 1024x1924 to 64x64.
4. Reduce the frame rate, e.g 60fps to 5fps.
5. Perform a higher level function to guess where the target area is with say a lower resolution, then perform the regular analysis on the cropped output, e.g. perform image recognition to locate the hand before determining the gesture.