Intelligent summarization of sports videos using automatic saliency detection

Abstract

The aim of this thesis is to present an efficient and intelligent way of creating sports summary videos by automatically identifying the highlights or salient events from one or multiple video footage using computer vision techniques and combining them to form a video summary of the game. The thesis presents a twofold solution -Identification of salient parts from single or multiple video footage of a certain sports event. -Remixing of video by extracting and merging various segments, with effects (such as slow replay) and mixing audio. This project involves applying methods of machine learning and computer vision to identify regions of interest in the video frames and detect action areas and scoring attempts. These methods were developed for the sport of basketball. However, the methods may be tweaked or enhanced for other sports such as football, hockey etc. For creating summary videos, various video processing techniques have been experimented to add certain visual effects to improve the quality of summary videos. The goal has been to deliver a fully automated, fast and robust system that could work with large high definition video files.