An object detection method for describing soccer games from video

Abstract

We propose a novel object detection and tracking method in order to detect and track objects necessary to describe contents of a soccer game. On the contrary to intensity oriented conventional object detection methods, the proposed method refers to color rarity and local edge property, and integrally evaluates them by a fuzzy function to achieve better detection quality. These image features were chosen considering the characteristics of soccer video images, that most non-object regions are roughly single colored (green) and most objects tend to have locally strong edges. We also propose a simple object tracking method, that could track objects with occlusion with other objects using a color based template matching. The result of an evaluation experiment applied to actual soccer video showed very high detection rate in detecting player regions without occlusion, and promising ability for regions with occlusion.