Video to Text Summarisation and Timestamp Generation to Detect Important Events

Abstract

With the advent of modern technology and the subsequent rise of efficient storage devices we are witnessing a rise in the number of media that is available to us. Among the most common media, the only one that takes up huge spaces on physical storage devices are videos. The primary reason for that is the addition of higher resolution videos and a greater frame rate. It is quite necessary to come up with summarisation techniques that help us understand the most important parts of the video. Apart from that, summarisation also helps us skip the non-essential parts of the video. This technology can be utilised to cut short on the time wasted on searching through the most relevant parts of the video. This paper tries to focus on the fundamental problem of summarising long videos and converting them into shorter sections that can effectively convey the same content if one were to see the entire video. Introducing timestamps also helps the viewer in jumping to the crucial events of the video. This paper makes use of deep learning algorithms such as Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN). These serve as a means of comparing different frames and generating end results.