Action recognition in free style wrestling using histogram of graph vertices from silhouette skeletons

Human Action and behavior recognition have many applications in computer vision and researchers have been working on this area for many years. Two-player sport action recognition is one of the research gaps in this scope. In this research, free style wrestling actions have been considered and by providing a dataset, an algorithm was developed to recognize such actions and different experiments were implemented. The free graph produced from player’s skeletons is used for feature extraction. In each frame, a feature vector is built using2-dimensional polar histogram of the graph points and by different combination of these vectors the final feature vector is produced for a video sample. Two classifiers; SVM and KNN were used independently to classify the actions based on different feature vector combinations. The highest score for action recognition is around%90 when KNN is used.