The goal of motion-based segmentation is to partition the image into regions that have different characteristics or properties. The paper establishes feasibility of using computer vision algorithms for real-time segmentation and compression of motion video sequences. A convex formulation, using Huber's regularizer, in a robust estimation framework has significant advantages over previous approaches. Unlike previous techniques, our approach guarantees stable, repeatable (or reproducible) segmentations which make real-time applications in segmenting video possible.