Skeletonization of Players in Dynamic Backgrounds Using Discrete Curve Evolution

摘要

Skeletal part extraction of the human or player becomes important in applications like developing the gaming consoles, event prediction in sports, gait based human recognition and classification of human activity etc. The efficiency of the said applications depend on how efficiently the skeletal part is extracted and the extraction of skeletal part is influenced by dynamic background of the video, occlusion and resolution of the video. So in this work, we proposed a method to extract the skeletal part of the sports man with a varying background thereby facilitating the subsequent analysis. Histogram of Oriented Gradients (HOG) is used to detect the human region by making use of Support Vector Machines (SVMs) and then Graph Cut technique is applied to remove the background to extract only the foreground in the form of silhouette. Finally a skeletal pruning method is applied which is based on contour partitioning method such as Discrete Curve Evaluation (DCE) technique. The proposed method is tested on sports video like cricket and shows effectiveness of the method in extracting the skeletal shape from the video with dynamic backgrounds.