Real-time video sequences matching using the spatio-temporal fingerprint
Ondrej Pribula, Jan Pohanka, Jan Fischer · 2010
With the growing popularity of digital multimedia distributed over broadband channels and networks copyright owners are requesting broadcasted content monitoring to keep their rights. A special need for multimedia monitoring is coming from advertisement submitters who would like to know how exactly their commercials are broadcasted and which categories of beholders are addressed. The basic unit needed for automated broadcast monitoring and its content evaluation is a robust and fast-enough video matching method. This paper proposes a method that can be used for matching of video streams or fast seeking of a short video sequence in a long video stream. According to real-time run requirements, the developed method is based on the brightness profile processing of significant image regions. Output from the algorithm is a one dimensional time-discretized vector of video sequence fingerprints that contains information about temporal and spatial video content. These fingerprints are appropriate to be stored in databases because of their significantly lower size in comparison to the source video. Having analyzed real video data, we show that reliable video sequence matching using introduced method can be achieved by using a video fingerprint data flow of 32 bytes per second of video stream. According to the robustness of proposed method this rate can be changed. This feature is very important for image processing systems with reduced data transfer bandwidth or low storage capacity, e.g. for the purpose of video-based TV channels rating measurement.