Development of motion capture system using multiple depth sensors

Abstract

The motion capture technology is used in various scenes. Several types of motion capturing systems (e.g., optical, inertial, magnetic) have been developed so far. These motion capture methods can get high precision motion capture data, but they need to attach markers or sensors on a human body. It may disturb the measurement of natural motion. There are marker less motion capture methods with RGB or depth image sensor, such as Microsoft Kinect. However, these methods cannot measure the motion in occlusion regions of sensor. In this paper, we propose the motion capture system using multiple depth sensors. At first, multiple depth data from the depth sensors are transformed into multiple point clouds, and they are combined into a combined point cloud. Then, the system computes skeleton lines from the combined point cloud by using the Reeb graph. Finally, the system estimates the joint positions from the skeleton lines according to the joint structure of the human. Through an experiment, this method can estimate accurate joint position when the object is standing with large angle to the sensor.