Algorithms Overview

Theory

The optic flow algorithm proposes by Srinivasan uses a reference image f₀ and shifted and rotated instances of the reference image f₁, f₂, f₃, f₄, f₅, f₆. It is assumed that the input image f can be interpolated as a linear combination of f₀ ... f₆. The motion from the input image to the reference image is found by determining the parameters of this linear combination giving a rotation and a translation. The solution is found by the least squares estimate

min ( f - (f₀ + 0.5 u (f₂ - f₁) + 0.5 v (f₄ - f₃) + 0.5 r (f₆ - f₅) ) ²

Srinivasan, M. V. (1994). An image-interpolation technique for the computation of optic flow and egomotion. Biological Cybernetics, 71(5), 401–415.

Algorithm

The sample implementation written in C can be found at

visual_processing/optic_flow/Srinivasan.cpp

The presented implementation uses fixed shift and rotation values. The mapping for the computation of the rotated images is precomputed during initialization of the method.