Kinect skeleton coordinate calibration for remote physical training

With the advent of the Microsoft Kinect sensor, skeleton coordinate systems have become an active part of interactive multimedia applications. The skeleton coordinate data captured by the Kinect sensor can be used to compare the similarity of remote users' motions in remote training systems. However, this approach is limited in that the remote users' initial positions have to be at the same position and face the sensor with the same angle. This paper proposes a Kinect Skeleton Coordinate Calibration (KSCC) algorithm to calibrate the remote user's arbitrary initial positions, thereby removing the above limitations on the initial positions and angles. It collects the remote user's initial position data, and calculates the initial centre coordinate of the user and initial angle between user and Kinect sensor. After the collection and calculation, all skeleton coordinates are transformed to a universal coordinate system according to the initial centre coordinate and rotated by a quaternion rotation. An evaluation test has been performed to assess the accuracy and limiting fields of the system. The results show that our approach is able to calibrate the Kinect skeleton coordinates with a high accuracy, with the requirements that the user's initial positions only need to be in the detection zone of the Kinect sensor.