A Geometrical Key-Frame Selection Method Exploiting Dominant Motion Estimation in Video

We describe an original method for selecting key frames to represent the content of every shot in a video. We aim at spatially sampling in an uniform way the coverage of the scene viewed in each shot. Our method exploits the computation of the dominant image motion (assumed to be due to the camera motion) and mainly relies on geometrical properties related to the incremental contribution of a frame in the considered shot. We also present a refinement of the proposed method to obtain a more accurate representation of the scene, but at the cost of a higher computation time, by considering the iterative minimization of an appropriate energy function. We report experimental results on sports videos and documentaries which demonstrate the accuracy and the efficiency of the proposed approach.