OPTICAL-FLOW FROM CONSTRAINT LINES PARAMETRIZATION

Optical flow is a measure of the change of image brightness in a frame sequence and is commonly used as an approximation of the velocity field, which is the perspective projection of the three-dimensional (3D) real velocity on the image plane. One method for evaluating optical flow is based on the Optical Flow Constraint (OFC) equation. Such an equation is associated with each image pixel. Under the assumption that in the immediate neighbourhood of a pixel the optical flow field is smooth, the constraint equations in that neighbourhood should have a common solution. Least-squares techniques are commonly used to derive the solution, which typically is very sensitive to discontinuities. A new method of evaluating the optical flow from the OFC equation is presented, which performs well in the presence of discontinuities. The method proposes to accumulate evidence from the solution of all pairs of constraints in each image segment to provide the most probable value for optical flow, rather than adopting least-squares techniques. The proposed accumulation of evidence is performed using a variant of the Hough transform, the Combinatorial Hough transform. Experimental results under different types of motion are presented. An efficient implementation of the algorithm is discussed on a highly parallel architecture.