Object detection by associating color frequency with local edge information and its application to player detection in soccer videos

摘要

This paper presents a new technique for detecting and tracking players in soccer image sequences. Generally, most object detection techniques, such as boundary-based approaches and region-based approaches, refer to intensity. Since when intensity is referred, foreground pixels tend to be absorbed by background pixels, object segmentation accuracy declines before further processes are performed. To overcome this problem, fuzzy evaluation function that associates color frequency with local edge information and performs region distinction per pixel is proposed. Using this function, object regions ignored by the intensity information is complemented. This technique is applied to the detection of the player region in soccer videos. We confirmed the validity of our method through an evaluation experiment.