I was writing a paper and needed some help. Can any explain why ORB feature is so much more computationally efficient than SURF and SIFT features?
The ORB uses a template-based corner detector that is named FAST. In the FAST corner detector, a decision tree is learned to determine the order of pixels in the circular template. Using a decision tree to determine a point as a corner accelerates the processing time. This is the most important part of the speed factor.
SIFT uses DoG pyramid and Hessian Matrix. Working with the DoG is time-consuming. However, SURF uses box filters for the Hessian Matrix. It is faster than the Difference of Gaussians but not as fast as ORB (that uses FAST corner detector). These two methods (SIFT and SURF) are based on the partial differentiation on Gaussian scale-spaces. Therefore, the ORB feature detection method is more computationally efficient than SIFT and SURF methods.