Robust Spatial Regularization and Velocity Layer Separation for Optical Flow Computation on Transparent Sequences

摘要

Motion estimation in sequences with transparencies is an important problem in robotics and medical imaging applications. In this work we propose two procedures to improve the transparent optical flow computation. We build from a variational approach for estimating multivalued velocity fields in transparent sequences. That method estimates multi-valued velocity fields which are not necessarily piecewise constant on a layer -each layer can evolve according to a non-parametric optical flow. First we introduce a robust statistical spatial interaction weight which allows to segment the multi-motion field. As result, our method is capable to recover the object's shape and the velocity field for each object with high accuracy. Second, we develop a procedure to separate the component layers of rigid objects from a transparent sequence. Such a separation is possible because of the high accuracy of the object's shape recovered from our transparent optical flow computation. Our proposal is robust to the presence of several objects in the same sequence as well as different velocities for the same object along the sequence. We show how our approach outperforms existing methods and we illustrate its capabilities on challenging sequences.