In daily life, stair walking is the basic motion and requires more dynamic motion than walking on level ground. On a stair’s environment, the most common cause of falling is due to ones’ posture, such as loss of balance. However, almost experimental equipment for motion capture which can collected the balance data such as “MVN”, a wearable motion capture solution with inertial sensors, are bulky and difficult to carry. For another, some solution focused on detecting and reconstructing 3D human motion from 2D content did not have enough accuracy.

Therefore, in this study we try to compare the similarities and differences between the body motion of ordinary people and elderly people when walking stairs at first, and then aim to provide a usable predication model for 2D content-based 3D motion data acquisition by using two types of motion capture devices and machine learning technology.

Based on our plan, the experiments have done on 2019.8, college students with different genders have been invited as the subjects to test their body motion when claiming stairs. Collected data are being analyzed and the results will be published in the international conference CyberSciTech2020.