Real time commercial detection using MPEG features

Advanced personal digital recorders (PDR) are already on the market. All of these devices offer time-shift, personalization, and retrieval features in addition to large storage capacity. Commercial detection is one of the most desirable features for these devices. Due to constraints on the consumer platforms we need to investigate content-based analysis techniques, which can use features, produced during MPEG compression. In this paper we describe commercial detection algorithm, implementation and results using compressed domain video analysis. We present the algorithm which uses features as triggers and verifiers to perform the commercial detection. We have achieved a recall of 93% and a precision of 95% when station logos and trailers are excluded and a precision of 99% when station l and trailers are regarded as a part of the commercial break.