In this paper, a new kind treadmill system based on Kinect is developed, and running action recognition algorithms are presented. This system is under the development of Unity game engine, 3DSMax, Maya, Photoshop modeling and graphic design software, it obtains various skeletal data of human running action by using Kinect- Microsoft somatosensory equipment and personal computer, skeletal binding based recognition algorithms of running, waving, jumping and squat were given under analysis of movement feature. Interactive running fitness entertainment sports can be implemented in this system through the recognition of human actions. The experimental data shows the effectiveness of action recognition algorithms based on skeletal binding, the treadmill system has the advantages with smaller volume hardware equipment, better software operating method by using human body action instead of mouse and keyboard, and also combines function with the running fitness and entertainment, replaces the traditional treadmill effect in some parts it has the very good somatosensory interaction effect and practical value, the research results can be used for more human-computer interaction application areas.