Research for human action recognition based on depth information

Action recognition is always a very active and challenging research topic in the field of computer vision. Kinect, as a motion sensing input device which can capture both color and depth images, open up new possibilities of dealing with this task. In this paper, we build a system to track and classify the human action based on the Microsoft Kinect technology. During the feature extraction step, we use the depth camera to capture depth information of human and propose a method to generate a skeleton model as our main features. Then, a multi-class Support Vector Machine (SVM) is adopted to recognize actions. Our method simplifies and systematizes the task of human action recognition meanwhile the experimental results show promising performance.