Abstract
Video classification is an important and active field of computer vision research due to its variety of applications (e.g. action recognition), ranging from intelligent surveillance to human-computer interaction. Recently, the rapid growth of camera devices and multimedia sharing platforms has created a tremendous amount of video data. Accordingly, there is the growing demand of the parallel and distributed video processing for dealing with the data-intensive and computing-intensive tasks of the recognition system. These tasks are widely recognized to be time-consuming when running over large-scale video data. In this paper, we propose a distributed framework built on the top of the emerging big data technology Spark which is well-designed for the in-memory parallel computing system. Particularly, based on the state-of-the-art bag of visual words model (BoV) for video classification, we provide a comprehensive study of key steps in BoV (i.e. feature extraction, codebook generation, and feature encoding), and propose their fast and distributed algorithms. In experiments, our results further demonstrate the efficiency of proposed algorithms as well as the scalability of our framework on a public dataset UCF101 by greatly speeding up the process of video classification.