People ROI compares and contrasts using image resizing and motion detection to obtain a ROI (region of interest) versus processing frames and regions unchanged. People detection is used in this example, but any type of image analysis can benefit from the techniques presented. Performance gains are in the 10 to 100 times range. My goal is to make people detection real-time on low end hardware without using hardware/software specific hacks (i.e. UncannyCV). The code is in Python and OpenCV 2.4.6.1 was used.
The wiki has installation instructions and techniques used.