Advertisement detection in TV recordings

Your goal is to develop a Linux software (console tool) which is able to detect advertisements in TV recordings. We will call it AdDetector. The software must be very fast and optimized. You will provide source codes with Makefile. After compiling, the result will be one binary file which will be called with one argument - path to the recording file for analysis. The format of video file is .TS (MPEG transport stream). The AdDetector will print on standard output where it detected advertisement start, and where it detected advertisement end. It must also print the exact file position (byte offset) where starts the keyframe in consideration. Example output: adstart 23s 323245b adend 220s 5357275b :: You can take into consideration this: - block of ads is usually surrounded by black frames. But it doesn't have to. - ads are usually more loud than the rest of the movie, but it doesn't have to. - ads usually exclude the TV station logo, but it doesn't have to. - ad block is always longer than 1 minute - ads have very high cut rate. But action movie can too. - the recording may start with an actual advertisement currently broatcasting, but it doesn't have to. Remember that you must optimize reading of the entire file, only read 1 keyframe per second or implement other techniques, because every recording is very big (several gigabytes) and it would cost lots of time to read all bytes. Similarly for audio. You can reuse any open source libraries or free sources which are available for Debian. Your result will be open source using GNU GPL license. Example TV recordings are below: [login to view URL] [login to view URL]