Content-based soccer video analysis, segmentation and summarization

Abstract

The project aims to develop a tool that can perform automatic analysis and summarization of soccer videos. Soccer is the most popular sport in the world. Many soccer videos are produced. Traditionally they are analyzed manually but it costs valuable time. Therefore it is necessary to have a tool that does the job automatically. The analysis includes some low-level processing algorithms such as grass field detection, shot boundary detection and shot classification, as well as a higher-level algorithm for penalty-box detection. First, the tool detects the grass field by finding the green pixels that form a large continuous region. Next, the tool detects the shot boundary by comparing the grass pixel ratios and similarity of color histograms of consecutive video frames. Then, the tool classifies the video shots by looking at the grass ratios of the frames in each video shot. After that, the program detects penalty-box by the three-parallel-line rule. Lastly, two types of summaries are produced. They are the soccer match play-break classification and the total time of offence of each team. The first type is based on the result of shot classification and the second one is based on the result of penalty-box detection. The effectiveness of the tool is demonstrated over soccer videos from different countries. The tool performs very well as its analysis results achieve high accuracy. Therefore it is possible letting the computer to analyze the soccer videos itself.