A single Kinect sensor could capture only an unobstructed human skeleton. In practice, two or more sensors are needed to accurately construct a complete skeleton of human in motion. However, using many sensors will present the problem of having multiple skeletons, each with different camera viewpoint. This work proposes the method of compositing skeleton from motion captures using multiple Kinect sensors in real time. Experiments show acceptable results. And the proposed method is promising for the future work in human motion captures.