Real-time monitoring of television advertisement using BoW

摘要

This paper presents a real-time monitoring of television advertisements using Bow. Frames are the basis of videos, thus an adaptive approach for extracting keyframes is proposed to describe video ads. Since in the BoW framework, the information of local feature points is lost during quantization, color name (CN) feature is introduced in this paper to enhance the discriminative power of visual words. Time consistent matching (TCM) filters matching frames that are not consistent in terms of order. Experiments show that our approach can monitor television advertisements accurately in real time.