AUTOMATED VIDEO CONTENT SUMMARIZATION

AUTOMATED VIDEO CONTENT SUMMARIZATION

Sinchana S, Sudarshan S Kakalwar, Vaibhav E S, Veekshith V
Under the guidance of  Prof.Belji T Associate Professor

Department of Computer Science and Engineering

Bachelor of Engineering

K. S. School of Engineering and Management, Bengaluru – 560109

ABSTRACT
. Automated video content summarization aims to reduce long videos into concise and straightforward summaries. Humans can recognize critical highlights in a video thanks to audio-visual and scenographic cues; however, the majority of methods do not cut it. They lack reliance on multi-modal features, and algorithmic appreciation, or disregard video aesthetics during editing, which compromises the quality of synthesis. In addition, many approaches rely heavily on labeled training data, which is not only tedious to compile, but also devoid of the vast content, and myriad of personal preferences encapsulated within different videos.

To overcome these problems, we propose ADUVS, an unsupervised, aesthetics-based automated video summarization framework. An aesthetics encoder sustains processing of visually important elements to identify the most captivating and salient video segments, including portions that are relevant. We also designed a multi-modal fusion module so different pieces of information can be combined to assess which fragments the summary should contain.

Of note, ADUVS is more economical and simpler to train thanks to its lack of need for annotated data. Through rigorous testing, our approach tested against numerous benchmark techniques and founds ADUVS surpassed them all based on evaluation metrics.