Domain-adaptive human motion and dynamics capture

该项目的目标是开发一种基于视频的无标记人体运动和动力学捕获系统。与其他运动捕捉(MOCAP)系统不同的是，该项目关注的是在传统的MOCAP摄影棚中难以记录的运动，从而导致缺乏特定运动域的准确3D数据。例如，这些动作可以是人类的户外活动，如攀岩、滑雪和跑酷，甚至可以是动物的动作。此外，为了估计被观测物体的准确3D重建，该系统还应该推断驱动运动的物理参数，例如关节扭矩和外力。这些力不仅对运动分析有很高的兴趣，而且还将改进视频的3D运动重建。此外，基于刚体动力学的约束将非常有助于在没有可用的地面真实3D记录的情况下实现对运动的域适应。

The goal of the project is to develop a video-based markerless human motion and dynamics capture system. In contrast to other motion capture (MoCap) systems this project focuses on motions that are hard to record in traditional MoCap studios resulting in a lack of accurate 3D data for the specific motion domain. For example these motions can be outdoor activities of humans such as rock climbing, skiing and parkour running or even motions of animals. Additionally to estimating accurate 3D reconstructions of the observed object the system should also infer physical parameters that drive the motion, for instance joint torques and external forces. These forces are not only of high interest for motion analysis but will also improve the 3D motion reconstruction from videos. Moreover, constraints based on rigid body dynamics will be extremely helpful to enable a domain adaptation to motions without available ground truth 3D recordings.