A Global Analysis of Optical Snow for Arbitrary Camera Motions

Abstract

Optical snow is a new category of motion for highly cluttered scenes in which no spatial continuity can be assumed. Since no smoothness constraint can be imposed on the velocity field, traditional optical flow methods can no longer be used. However, a model of optical snow has recently been proposed and algorithms based on this model were suggested using an analysis in the spatio-temporal frequency domain. This model assumes lateral motion and can be used to solve the 3D camera motion problem by decomposing sequences in sufficiently small patches. We would like to use the same model to find arbitrary camera motions globally instead of using patches. In the present paper, we introduce a complementary model for purely non-lateral optical snow. The standard optical snow model and this complementary form could lead to a new global approach for solving the general egomotion problem. We show how non-lateral optical snow sequences can be rectified such that standard methods to analyze optical snow can be applied. The effectiveness of the method is shown for both real and synthetic sequences.

Documents

PDF

---