Body Parts Estimation for Motion Capture System Using Multiple Depth Sensors

Abstract

Capturing the motion is important in various scenes. Several types of motion capture systems such as optical or inertial type have been developed so far. These type motion capturing systems can capture the high precise motion. But these type motion capture systems need to attach the markers or some devices and disturb the natural motion. On the other hand, there are marker less motion capture methods by using RGB or depth sensor such as Microsoft Kinect. These methods can get the motion capture data without attachment of devices, but these methods cannot get the motion capture data in occlusion regions of sensor. To resolve the problem, we have developed a motion capture system with multiple depth sensors. At first, the system gets the multiple depth data and combines them by unifying the coordinate systems. Then, the system estimates which part of the body each point in the combined point cloud by the Random Forest. Finally, the system estimates the joint positions from the body parts. This paper proposes the method of body parts estimation for the proposed motion capture system and compares the accuracy of estimation among the features for Random Forests.