INCORRECT SKELETAL PARTS RECOVERING METHOD USING 3D SPACE DIVISION AND MOCAP DATA

摘要

Over the past few decades many researchers have been studying about human motion recognition. Human motion recognition can apply a various application in real world. Recently, the acquirement of the skeleton has been become a more easy with the release of Kinect which offer the estimated skeleton without marker. At the same time, the skeleton based human motion recognition and description method are proposed as a various method. However, this skeleton information is easy to lose the target joint and recognize the incorrect joint position information. In contrast, motion capture system can create a motion capture data which is including more natural and correct joint position. But the task to create a motion capture (MoCap) data is time consuming and labor intensive. Thus, we propose the application to offer recovered motion without MoCap data and recovering method using 3D space division and MoCap data. We extract a sequence of the joint trajectory by using 3D motion space. And we use a dynamic time warping algorithm (DTW) to find a best estimate joint in Kinect skeleton and a corresponding MoCap frame. Then, we repair the incorrect Kinect joint position based on corresponding MoCap data.