This work presents a novel approach for robust PCA with total variationregularization for foreground-background separation and denoising on noisy,moving camera video. Our proposed algorithm registers the raw (possiblycorrupted) frames of a video and then jointly processes the registered framesto produce a decomposition of the scene into a low-rank background componentthat captures the static components of the scene, a smooth foreground componentthat captures the dynamic components of the scene, and a sparse component thatcan isolate corruptions and other non-idealities. Unlike existing methods, ourproposed algorithm produces a panoramic low-rank component that spans theentire field of view, automatically stitching together corrupted data frompartially overlapping scenes. The low-rank portion of our robust PCA model isbased on a recently discovered optimal low-rank matrix estimator (OptShrink)that requires no parameter tuning. We demonstrate the performance of ouralgorithm on both static and moving camera videos corrupted by noise andoutliers.
展开▼
