Recognition of tennis actions using a depth camera

Human actions recognition has been one of the most popular subject areas in computer vision. Recently, the usage of depth cameras which are capable of generating three dimensional data enabled more complex human actions to be recognized. In this study, the problem of tennis actions recognition using a depth camera is tackled and a three dimensional tennis actions dataset has been created. To be able to recognize each tennis action, each image is represented with the three dimensional skeletal based features. Each tennis action sample is represented by appending the features of each image residing in the signature subset created with the k-means clustering method in a time based manner. With the help of supervised multi-class support vector machine method, tennis actions have been modeled with a remarkable success.