Current Motion Tracking from Satellite Image Sequence With Global Similarity Optimization Model

Abstract

A new approach is proposed for motion tracking from a satellite image sequence to address the issue of radiometric variations between two-frame images. A global similarity criterion is defined based on the cross correlation between two images to convert a convex optimization model to a nonconvex one. The retrieval of the motion field with the criterion of maximum similarity becomes solving a nonlinear minimization problem. One of the generic iterative equations is formulated based on the global similarity optimization model (GSOM) and a unified adaptive framework. The simplified iterative equation is easy to implement and highly efficient with lower computational complexity for motion estimation. The new GSOM method can adapt to the violations of the tracer conservation constraint for motion field estimation when there are radiometric variations between two images. The approach is tested using an ocean simulation data set and realistic satellite infrared image sequences. Experimental results indicate that the new approach is not only robust for radiometric variations between two images but also efficient, fast, and accurate for motion estimation.