Television Commercial Detection

The following pages summarize the Winter 2013 EECS 452 - Digital Signal Processing Laboratory Senior Design Group: Television Commercial Detection which consisted of:
Hassan Hamid, Robert Klosek, John Marcoux, Siddharth Sundar, and Garrett Swaney

Introduction and Motivation

Our project aims to create a system that is capable of detecting commercial transitions in real-time and allows users to avoid watching commercials, and return back to their television program. Television is widespread in American culture and offering the opportunity to avoid the aggravation and time wasted during commercials would change the way we watch TV. According to the 2012 Nielson Report, Americans watch on average thirty-four hours of TV each week. With each hour of programming containing around sixteen minutes of commercials, the average American watches about nine hours of commercials in a week (Hinckley). Our project allows users to watch a primary channel and switch to a secondary channel upon commercial detection, automatically returning back to the primary channel when the commercial is over.

Our idea for this project was validated by the following references: SVM Tutorial, RPi Archives, Matlab's HoG, HoG Description, and HoG for Human Detection. We use the channel logo as a method of determining whether the current frame is part of a television show or commercial. This involves cropping out the logo portion of the image and performing a classification algorithm on it. We used the Histogram of Oriented Gradients feature descriptor presented here. Finally we created a Support Vector Machine to determine the classification of the input test vector.