Pseudo Fuzzy Colour Calibration for Sport Video Segmentation

Abstract

Video segmentation is one of the most important parts of a vision system which allows partitioning each frame into homogeneous regions that share a common property. This work proposes a new methodology that aggregates three different techniques: background subtraction, region growing and a pseudo Fuzzy colour model to define colour subspaces that characterize each class. In addition, the pseudo Fuzzy colour model allows a given colour to belong to more than one class and enables the expansion of the classes through a dynamic model based on belonging and persistence information. In case of shared colours among classes, regional features are searched in order to determine the object's class. Tests with test and real videos of sports footages show promising results.