Computer vision applications to the study of sea-ice motion in Antarctica

The motion of sea-ice in Antarctica is studied using QuikSCAT scatterometer imagery using methods from computer vision and image processing such as intensity edges and optical flow (OF). Features are computed as differential invariants based on spatial and temporal derivatives at various scales. The first estimates of the motion vector field obtained through optical flow are used as the starting point for a regularization scheme that imposes constraints that bring the estimate closer to feature tracking results and observed motion. Constraints from fluid dynamics are brought in by separating the motion field into its divergence free and rotational free components with another field obtained through a convex combination of these components. The advantages of this approach are that it produces a dense motion field that can be globally processed and locally adjusted to fit data and a model. This study complements the analysis of sea ice motion by application of wavelet theory.