A novel vision-based human motion capture system using dual-Kinect

In this paper, we present a novel vision-based human motion capture system, which employs two simultaneously running Kinects to scan the same subject in order to improve the stability. First the causes of the noise and the unstability of single Kinect system is analyzed. Then we introduce a very simple but effective method to calibrate two Kinects. As the Kinect´s depth sensing is based on the emission of an infrared pattern, two Kinects should be placed with a 180° angle to each other to reduce the interference errors caused by overlapping fields of view. Two Kinects capturing motions of the same subject results in two sets of data, each representing the motions from a certain angle. We propose a fusion algorithm with the redundant data to make up for the obstructed parts without doing any post estimation or approximation and stabilize the system as well. The experiment validates the effectiveness of the proposed approach for overcoming the noise and the occlusion problem.