Recognition of human actions by using depth information

Usage of 3. dimension information obtained from depth sensors in human action recognition become important recently. Depth information can increase recognition accuracy in some applications. In this study, 10 different human actions are tried to recognize on a human model derived from Microsoft Kinect RGBD sensor. Angles between joints and displacement of joints on 3 koordinat axes are used as features. Actions are classified with the random forest and support vector machine approaches and 96% classification accuracy is obtained with the random forest approach.