Advertisement Detection in Broadcast Videos

Abstract

Advertisement detection in videos has remained interesting area of research for the past three decades. For this purpose many techniques have been used. The focus of this study is to detect advertisement in broadcast videos using deep learning convolutional network. A 3D convolutional neural network has been used for this purpose. For this purpose conv3D has been used for the feature extraction of chunks of videos. First Spatio-temporal features are learned through C3D feature extractor and then the features are passed to multiple layer perceptron (MLP) which classifies the chunk as advertisement or non advertisement. The accuracy of the model is quiet promising.