Video scene detection is the task of dividing a video into semantic sections. I will present our novel and effective method for temporal grouping of scenes using an arbitrary set of features computed from the video. This task is formulated as a general optimization problem and an efficient solution is provided using dynamic programming. Our unique formulation allows us to directly obtain a temporally consistent segmentation, unlike many existing methods, and has the advantage of being parameter-free. I will also present how we expanded the method to incorporate features from multiple modalities, and I will present a novel technique to estimate the number of scenes in the video using Singular Value Decomposition (SVD) as a low-rank approximation of a distance matrix. This method proved to perform outstandingly and resulted in three published papers.
Daniel Rotman is a research scientist in the Video and GIS Analytics group at IBM Research - Haifa. Daniel received his B.Sc and M.Sc. at the Technion in the Electrical Engineering department.