Applying approximate computing principles for real-time rolling shake hyper-stabilization during video recording

The problem of video stabilization is essentially a composition of three problems: Determining the magnitude and direction of movement of the camera; Differentiating intentional camera motion from unwanted camera shake; Estimating the magnitude and direction of the unwanted camera shake and correcting for it. The principles of approximate computing present a beautiful solution to these problems. It enables real-time stabilization, even hyper-stabilization (near absolute stabilization) of camera rolling shake. In this short paper, we discuss the technical details of this concept and how approximate computing philosophy is an integral part of its design and implementation. We also look at its shortcomings and some powerful improvements to mitigate them by again applying the principles of approximate computing.