Estimation of optical flow by voting using weight function

摘要

Estimation of optical flow is an available method for object motion analysis. However, there is a problem such that optical flows are not estimated precisely by the influence of noise. For a solution of the problem, the method using voting process that have robustness against noise have been proposed. However, in the situations that object motions are complex such as expansion, reduction or rotation, there is a problem that optical flows are not estimated precisely by decrease of reliability of maximum voting value. In order to solve the problem, we suggest a method that reevaluate a density of voting values by a convolution of voting value using an weight function.