Determining motion saliency in videos

Abstract

This Final Year Project describes an algorithm for obtaining and extracting information that attracts user attention in a video clip. The key factors that make a region attractive are the motion that happened in a video, the low-level features like image orientation and intensity and also the semantic influence in a video.

The exploration of the project is supported by research and study on several related works. The function of analyzing the related work is to aid in the project planning cycle. With this workflow, it will be able to help in project designing and also prepare for project implementation.

The purpose of this project is to create an algorithm that identifies the attention attracting regions and generate a skimmed version which refers to a concise representation of the original video clip. The region of attractions in a video clip in this project is identified by motion. The motion saliency information is generated based on motion vector information with spatial and temporal coherency. During the process, experimental results were obtained and evaluated to justify this accuracy. The final results of the skimmed video were produced and it showed excellent results.